From cc6dcecfc837e09b22706325d003b8d420260a32 Mon Sep 17 00:00:00 2001 From: DWarez Date: Thu, 25 Jun 2026 18:19:29 +0200 Subject: [PATCH 1/5] add: tutorial for GRPO on Sagemaker using TRL Signed-off-by: DWarez --- .../grpo-llm-trl/sagemaker-notebook.ipynb | 991 ++++++++++++++++++ 1 file changed, 991 insertions(+) create mode 100644 docs/sagemaker/notebooks/sagemaker-sdk/grpo-llm-trl/sagemaker-notebook.ipynb diff --git a/docs/sagemaker/notebooks/sagemaker-sdk/grpo-llm-trl/sagemaker-notebook.ipynb b/docs/sagemaker/notebooks/sagemaker-sdk/grpo-llm-trl/sagemaker-notebook.ipynb new file mode 100644 index 000000000..bf35d8e06 --- /dev/null +++ b/docs/sagemaker/notebooks/sagemaker-sdk/grpo-llm-trl/sagemaker-notebook.ipynb @@ -0,0 +1,991 @@ +{ + "cells": [ + { + "cell_type": "markdown", + "id": "b1374ffe", + "metadata": {}, + "source": [ + "# Improve tool calling with GRPO on Amazon SageMaker\n", + "\n", + "In this notebook, you'll fine-tune **`HuggingFaceTB/SmolLM3-3B`** to produce better single-turn tool calls. The training signal is fully verifiable: instead of asking another model to judge the answer, we compare the model's generated tool call with the ground-truth call from the dataset.\n", + "\n", + "We'll use GRPO from TRL because it can optimize directly from a Python reward function. For each prompt, the model samples several completions, the reward function scores each completion, and GRPO updates the model toward the completions that scored better than the rest of the group.\n", + "\n", + "You'll walk through the full SageMaker training path:\n", + "\n", + "- prepare a function-calling dataset for GRPO,\n", + "- write an exact-match reward for tool name and arguments,\n", + "- add a lightweight format reward for well-formed `` JSON,\n", + "- package the training script SageMaker will run,\n", + "- launch a multi-GPU training job,\n", + "- plot the reward curves from the training logs.\n", + "\n", + "You need AWS credentials, a SageMaker execution role with S3 access, quota for `ml.g6.12xlarge` training in `us-east-1`, and a Hugging Face token that can download the base model.\n" + ] + }, + { + "cell_type": "markdown", + "id": "5b77b8dd", + "metadata": {}, + "source": [ + "## Setup\n", + "\n", + "Install only the packages used by the notebook kernel. The training dependencies are already inside the container that SageMaker runs.\n", + "\n", + "This notebook uses SageMaker SDK v3 (`ModelTrainer`). The older v2 `HuggingFace` estimator uses different launch objects.\n" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "6fb84692", + "metadata": {}, + "outputs": [], + "source": [ + "%pip install -q \"sagemaker>=3\" datasets" + ] + }, + { + "cell_type": "markdown", + "id": "ecc3556e", + "metadata": {}, + "source": [ + "The training job needs:\n", + "\n", + "- `HF_TOKEN` in the container environment, so the model download can authenticate with the Hub.\n", + "- A SageMaker execution role, so SageMaker can pull the image, read the S3 input channel, and write the model artifact.\n", + "\n", + "If `get_execution_role()` cannot find a role, replace the placeholder ARN with a role SageMaker can assume.\n" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "e1a0499a", + "metadata": {}, + "outputs": [], + "source": [ + "import os\n", + "from huggingface_hub import login, get_token\n", + "\n", + "# Uses the HF_TOKEN env var if set; otherwise opens an interactive login prompt.\n", + "if not os.environ.get(\"HF_TOKEN\"):\n", + " login()\n", + "HF_TOKEN = get_token()\n", + "assert HF_TOKEN, \"No HF token found — set HF_TOKEN or run login()\"\n", + "print(\"HF token loaded\")" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "954730c2", + "metadata": {}, + "outputs": [], + "source": [ + "import boto3\n", + "from sagemaker.core.helper.session_helper import Session, get_execution_role\n", + "\n", + "# The training image lives in us-east-1, so keep the job, the S3 bucket and the image in one region.\n", + "REGION = \"us-east-1\"\n", + "sess = Session(boto_session=boto3.Session(region_name=REGION))\n", + "bucket = sess.default_bucket()\n", + "\n", + "try:\n", + " role = get_execution_role(sagemaker_session=sess)\n", + "except Exception:\n", + " role = \"arn:aws:iam:::role/\" # set this when running outside SageMaker\n", + "\n", + "print(\"region:\", REGION)\n", + "print(\"bucket:\", bucket)\n", + "print(\"role: \", role)" + ] + }, + { + "cell_type": "markdown", + "id": "38d2e42a", + "metadata": {}, + "source": [ + "## Configuration\n", + "\n", + "`MODEL_ID` is the base model. `INSTANCE_TYPE` controls the training hardware. `TRAINING_IMAGE` is the ECR image SageMaker runs.\n", + "\n", + "Keep `TRAINING_IMAGE`, the SageMaker job, and the S3 bucket in the same AWS region.\n" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "7280ac2e", + "metadata": {}, + "outputs": [], + "source": [ + "MODEL_ID = \"HuggingFaceTB/SmolLM3-3B\" # SmolLM3, HF's 3B instruct model\n", + "INSTANCE_TYPE = \"ml.g6.12xlarge\" # 4xL4 24GB\n", + "\n", + "TRAINING_IMAGE = \"754289655784.dkr.ecr.us-east-1.amazonaws.com/hf-trl-grpo:sagemaker-trl-dev-e63f67e\"" + ] + }, + { + "cell_type": "markdown", + "id": "853d6685", + "metadata": {}, + "source": [ + "## Prepare the dataset\n", + "\n", + "`Salesforce/xlam-function-calling-60k` has the three fields needed for a verifiable tool-call reward:\n", + "\n", + "- `query`: the user request.\n", + "- `tools`: the functions the model may call.\n", + "- `answers`: the correct tool call, including arguments." + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "2690d04e", + "metadata": {}, + "outputs": [], + "source": [ + "import json\n", + "from datasets import load_dataset\n", + "\n", + "raw = load_dataset(\"Salesforce/xlam-function-calling-60k\", split=\"train\")\n", + "print(raw)\n", + "print(json.dumps({k: raw[0][k] for k in (\"query\", \"tools\", \"answers\")}, indent=2)[:1200])" + ] + }, + { + "cell_type": "markdown", + "id": "e35dfc13", + "metadata": {}, + "source": [ + "`GRPOTrainer` reads prompts from a `prompt` column. The system message contains the tool list and the required `` format; the user message contains the request.\n", + "\n", + "Keep `answers` as a separate column. TRL passes extra dataset columns to the reward function as keyword arguments, so `answers` remains hidden from the model but available to the scorer.\n" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "2b17c98c", + "metadata": {}, + "outputs": [], + "source": [ + "import shutil\n", + "from pathlib import Path\n", + "\n", + "N_TRAIN = 2000 # a small subset keeps this demo cheap; the full set is 60k\n", + "\n", + "\n", + "def has_one_answer(row):\n", + " try:\n", + " answers = json.loads(row[\"answers\"])\n", + " except json.JSONDecodeError:\n", + " return False\n", + " return isinstance(answers, list) and len(answers) == 1\n", + "\n", + "\n", + "SYSTEM_TMPL = \"\"\"/no_think\n", + "You are an expert in composing function calls. Return exactly one function call that answers the user's request.\n", + "\n", + "You have access to the following tools:\n", + "\n", + "{tools}\n", + "\n", + "\n", + "Return a single JSON object with the function name and arguments within XML tags:\n", + "\n", + "{{\"name\": , \"arguments\": }}\n", + "\n", + "Emit exactly one block and no other text.\"\"\"\n", + "\n", + "\n", + "def to_grpo(row):\n", + " return {\n", + " \"prompt\": [\n", + " {\"role\": \"system\", \"content\": SYSTEM_TMPL.format(tools=row[\"tools\"])},\n", + " {\"role\": \"user\", \"content\": row[\"query\"]},\n", + " ],\n", + " \"answers\": row[\"answers\"],\n", + " }\n", + "\n", + "\n", + "single_call = raw.filter(has_one_answer)\n", + "train_source = single_call.shuffle(seed=42).select(range(min(N_TRAIN, len(single_call))))\n", + "prepared = train_source.map(to_grpo, remove_columns=raw.column_names)\n", + "if Path(\"prepared\").exists():\n", + " shutil.rmtree(\"prepared\")\n", + "prepared.save_to_disk(\"prepared\")\n", + "prepared\n" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "c15e88de", + "metadata": {}, + "outputs": [], + "source": [ + "print(json.dumps(prepared[0][\"prompt\"], indent=2)[:1500])\n", + "print(\"\\nground truth:\", prepared[0][\"answers\"])" + ] + }, + { + "cell_type": "markdown", + "id": "4d27f7be", + "metadata": {}, + "source": [ + "Upload the prepared dataset to S3. The input channel is named `train`, so SageMaker mounts it at `$SM_CHANNEL_TRAIN`; `train.py` loads the dataset from that path.\n" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "0859c513", + "metadata": {}, + "outputs": [], + "source": [ + "train_s3 = sess.upload_data(\"prepared\", bucket=bucket, key_prefix=\"xlam-grpo/prepared\")\n", + "print(train_s3)" + ] + }, + { + "cell_type": "markdown", + "id": "02dafe46", + "metadata": {}, + "source": [ + "## The reward function\n", + "\n", + "The reward must not crash on bad model output. Bad JSON, missing tags, extra prose, or wrong arguments return a low score.\n", + "\n", + "- `tool_call_reward`: exact match on tool name and arguments.\n", + "- `format_reward`: partial credit for valid `` JSON, useful before exact matches are common.\n", + "\n", + "Both functions are written to `src/rewards.py` so the training job can import them.\n" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "8dab8b5e", + "metadata": {}, + "outputs": [], + "source": [ + "from pathlib import Path\n", + "Path(\"src\").mkdir(exist_ok=True)" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "11c358f0", + "metadata": {}, + "outputs": [], + "source": [ + "%%writefile src/rewards.py\n", + "'''Verifiable rewards for GRPO tool-call training.\n", + "\n", + "Parse the tool calls the model emits (`` format) and compare them\n", + "against the dataset's ground-truth `answers`. No LLM judge, no sandbox.\n", + "'''\n", + "import json\n", + "import re\n", + "\n", + "_TOOL_CALL_RE = re.compile(r\"\\s*(.*?)\\s*\", re.DOTALL)\n", + "\n", + "\n", + "def _text(completion):\n", + " # GRPO hands completions as a string or a list of chat messages; return the text.\n", + " if isinstance(completion, str):\n", + " return completion\n", + " if isinstance(completion, list):\n", + " for msg in reversed(completion):\n", + " if isinstance(msg, dict) and msg.get(\"role\") == \"assistant\":\n", + " return msg.get(\"content\") or \"\"\n", + " return str(completion)\n", + "\n", + "\n", + "def parse_tool_calls(text):\n", + " '''Extract {name, arguments} dicts from a completion. Never raises.\n", + "\n", + " If tags are used, reject any non-whitespace text outside them.\n", + " '''\n", + " text = _text(text)\n", + " matches = list(_TOOL_CALL_RE.finditer(text))\n", + " if matches:\n", + " outside = _TOOL_CALL_RE.sub(\"\", text)\n", + " if outside.strip():\n", + " return []\n", + " blocks = [m.group(1) for m in matches]\n", + " else:\n", + " blocks = [text.strip()] # fall back to bare JSON\n", + " calls = []\n", + " for block in blocks:\n", + " try:\n", + " obj = json.loads(block)\n", + " except (json.JSONDecodeError, TypeError):\n", + " continue\n", + " for call in obj if isinstance(obj, list) else [obj]:\n", + " if isinstance(call, dict) and \"name\" in call:\n", + " calls.append({\"name\": call[\"name\"], \"arguments\": call.get(\"arguments\", {})})\n", + " return calls\n", + "\n", + "\n", + "def _loose_tool_calls(text):\n", + " '''Parse tagged calls even when the completion rambles outside the tags.'''\n", + " text = _text(text)\n", + " calls = []\n", + " for match in _TOOL_CALL_RE.finditer(text):\n", + " try:\n", + " obj = json.loads(match.group(1))\n", + " except (json.JSONDecodeError, TypeError):\n", + " continue\n", + " for call in obj if isinstance(obj, list) else [obj]:\n", + " if isinstance(call, dict) and \"name\" in call:\n", + " calls.append(call)\n", + " return calls\n", + "\n", + "\n", + "def _key(call):\n", + " return json.dumps({\"name\": call.get(\"name\"), \"arguments\": call.get(\"arguments\", {})},\n", + " sort_keys=True, ensure_ascii=False)\n", + "\n", + "\n", + "def _normalize(calls):\n", + " return sorted(_key(c) for c in calls) # order- and key-order-independent\n", + "\n", + "\n", + "def _ground_truth(entry):\n", + " if isinstance(entry, str):\n", + " try:\n", + " entry = json.loads(entry)\n", + " except json.JSONDecodeError:\n", + " return []\n", + " return entry if isinstance(entry, list) else []\n", + "\n", + "\n", + "def tool_call_reward(completions, answers=None, **kwargs):\n", + " '''1.0 when the emitted tool calls exactly match the ground truth, else 0.0.'''\n", + " answers = answers or [None] * len(completions)\n", + " out = []\n", + " for completion, gt in zip(completions, answers):\n", + " pred = _normalize(parse_tool_calls(completion))\n", + " out.append(1.0 if pred and pred == _normalize(_ground_truth(gt)) else 0.0)\n", + " return out\n", + "\n", + "\n", + "def format_reward(completions, **kwargs):\n", + " '''Dense formatting signal: clean call wins; rambling/long junk loses.'''\n", + " scores = []\n", + " for completion in completions:\n", + " text = _text(completion).strip()\n", + " penalty = min(len(text), 2000) / 20000.0\n", + " if parse_tool_calls(text):\n", + " scores.append(1.0)\n", + " elif _loose_tool_calls(text):\n", + " scores.append(0.2 - penalty)\n", + " elif \"{\"name\": \"x\", \"arguments\": {}}'], answers=[gt])) # [0.0]" + ] + }, + { + "cell_type": "markdown", + "id": "33b36f36", + "metadata": {}, + "source": [ + "## The training script\n", + "\n", + "`train.py` reads SageMaker paths and CLI hyperparameters, builds `GRPOConfig`, trains, and saves to `$SM_MODEL_DIR`.\n", + "\n", + "Two generation settings are deliberate:\n", + "\n", + "- `renormalize_logits=True` makes sampling robust if a logits processor creates invalid probabilities.\n", + "- `attn_implementation=\"sdpa\"` uses the stable attention path for this stack.\n", + "\n", + "Generation uses the Transformers backend.\n" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "f42a218b", + "metadata": {}, + "outputs": [], + "source": [ + "%%writefile src/train.py\n", + "'''SageMaker entry script: GRPO tool-call training with TRL.\n", + "\n", + "Runs once per GPU under torchrun. Reads data from $SM_CHANNEL_TRAIN, writes the\n", + "model to $SM_MODEL_DIR. Hyperparameters arrive as --key value CLI args.\n", + "'''\n", + "import argparse\n", + "import json\n", + "import os\n", + "import shutil\n", + "\n", + "from datasets import load_from_disk\n", + "from transformers import AutoTokenizer, TrainerCallback\n", + "from trl import GRPOConfig, GRPOTrainer\n", + "\n", + "from rewards import format_reward, tool_call_reward\n", + "\n", + "\n", + "def str2bool(v):\n", + " return str(v).lower() in (\"1\", \"true\", \"yes\")\n", + "\n", + "\n", + "class JsonlLogCallback(TrainerCallback):\n", + " def __init__(self, paths, stop_on_collapse=False):\n", + " self.paths = paths\n", + " self.stop_on_collapse = stop_on_collapse\n", + "\n", + " def on_log(self, args, state, control, logs=None, **kwargs):\n", + " if not logs:\n", + " return\n", + " row = {\"step\": state.global_step, **logs}\n", + " collapsed = (\n", + " self.stop_on_collapse\n", + " and state.global_step >= 2\n", + " and (logs.get(\"completions/clipped_ratio\") or 0) >= 0.9\n", + " and (logs.get(\"rewards/tool_call_reward/mean\") or 0) <= 0\n", + " )\n", + " if collapsed:\n", + " row[\"early_stop_reason\"] = \"collapse: clipped completions with zero exact-match reward\"\n", + " control.should_training_stop = True\n", + " if state.is_world_process_zero:\n", + " for path in self.paths:\n", + " os.makedirs(os.path.dirname(path), exist_ok=True)\n", + " with open(path, \"a\", encoding=\"utf-8\") as f:\n", + " f.write(json.dumps(row) + \"\\n\")\n", + " return control\n", + "\n", + "\n", + "def parse_args():\n", + " p = argparse.ArgumentParser()\n", + " p.add_argument(\"--model_id\", default=\"HuggingFaceTB/SmolLM3-3B\")\n", + " p.add_argument(\"--train_dir\", default=os.environ.get(\"SM_CHANNEL_TRAIN\", \"/opt/ml/input/data/train\"))\n", + " p.add_argument(\"--output_dir\", default=os.environ.get(\"SM_MODEL_DIR\", \"/opt/ml/model\"))\n", + " # global batch (per_device * num_gpus * grad_accum) must be divisible by num_generations\n", + " p.add_argument(\"--num_generations\", type=int, default=8)\n", + " p.add_argument(\"--per_device_train_batch_size\", type=int, default=8)\n", + " p.add_argument(\"--gradient_accumulation_steps\", type=int, default=4)\n", + " p.add_argument(\"--learning_rate\", type=float, default=1e-8)\n", + " p.add_argument(\"--beta\", type=float, default=0.1)\n", + " p.add_argument(\"--temperature\", type=float, default=0.7)\n", + " p.add_argument(\"--top_p\", type=float, default=1.0)\n", + " p.add_argument(\"--top_k\", type=int, default=0)\n", + " p.add_argument(\"--min_p\", type=float, default=0.0)\n", + " p.add_argument(\"--max_grad_norm\", type=float, default=0.05)\n", + " p.add_argument(\"--max_completion_length\", type=int, default=128)\n", + " p.add_argument(\"--max_steps\", type=int, default=50)\n", + " p.add_argument(\"--reward_weights\", default=\"1.0,0.5\")\n", + " p.add_argument(\"--repetition_penalty\", type=float, default=1.05)\n", + " p.add_argument(\"--gradient_checkpointing\", type=str2bool, default=True)\n", + " p.add_argument(\"--deepspeed\", default=\"\")\n", + " p.add_argument(\"--report_to\", default=\"none\")\n", + " p.add_argument(\"--log_completions\", type=str2bool, default=True)\n", + " p.add_argument(\"--num_completions_to_print\", type=int, default=8)\n", + " p.add_argument(\"--stop_on_collapse\", type=str2bool, default=True)\n", + " p.add_argument(\"--remove_invalid_values\", type=str2bool, default=False)\n", + " p.add_argument(\"--renormalize_logits\", type=str2bool, default=True)\n", + " p.add_argument(\"--seed\", type=int, default=42)\n", + " return p.parse_args()\n", + "\n", + "\n", + "def main():\n", + " args = parse_args()\n", + "\n", + " tokenizer = AutoTokenizer.from_pretrained(args.model_id)\n", + " if tokenizer.pad_token is None:\n", + " tokenizer.pad_token = tokenizer.eos_token\n", + "\n", + " train_dataset = load_from_disk(args.train_dir)\n", + "\n", + " generation_kwargs = {\n", + " \"remove_invalid_values\": args.remove_invalid_values,\n", + " \"renormalize_logits\": args.renormalize_logits,\n", + " \"eos_token_id\": tokenizer.eos_token_id,\n", + " \"pad_token_id\": tokenizer.pad_token_id,\n", + " }\n", + " if args.min_p > 0:\n", + " generation_kwargs[\"min_p\"] = args.min_p\n", + "\n", + " config = GRPOConfig(\n", + " output_dir=args.output_dir,\n", + " per_device_train_batch_size=args.per_device_train_batch_size,\n", + " gradient_accumulation_steps=args.gradient_accumulation_steps,\n", + " learning_rate=args.learning_rate,\n", + " beta=args.beta,\n", + " temperature=args.temperature,\n", + " top_p=args.top_p,\n", + " top_k=args.top_k,\n", + " num_generations=args.num_generations,\n", + " max_completion_length=args.max_completion_length,\n", + " max_steps=args.max_steps,\n", + " max_grad_norm=args.max_grad_norm,\n", + " bf16=True,\n", + " repetition_penalty=args.repetition_penalty,\n", + " gradient_checkpointing=args.gradient_checkpointing,\n", + " gradient_checkpointing_kwargs={\"use_reentrant\": False},\n", + " generation_kwargs=generation_kwargs,\n", + " loss_type=\"dapo\",\n", + " mask_truncated_completions=False,\n", + " scale_rewards=\"batch\",\n", + " reward_weights=[float(w) for w in args.reward_weights.split(\",\")],\n", + " model_init_kwargs={\"dtype\": \"bfloat16\", \"attn_implementation\": \"sdpa\"},\n", + " deepspeed=args.deepspeed or None,\n", + " logging_steps=1,\n", + " log_completions=args.log_completions,\n", + " num_completions_to_print=args.num_completions_to_print,\n", + " save_strategy=\"no\",\n", + " report_to=args.report_to,\n", + " seed=args.seed,\n", + " )\n", + "\n", + " metrics_paths = [os.path.join(args.output_dir, \"training_metrics.jsonl\")]\n", + " if os.environ.get(\"SM_OUTPUT_DATA_DIR\"):\n", + " metrics_paths.append(os.path.join(os.environ[\"SM_OUTPUT_DATA_DIR\"], \"training_metrics.jsonl\"))\n", + "\n", + " trainer = GRPOTrainer(\n", + " model=args.model_id,\n", + " args=config,\n", + " reward_funcs=[tool_call_reward, format_reward],\n", + " train_dataset=train_dataset,\n", + " processing_class=tokenizer,\n", + " callbacks=[JsonlLogCallback(metrics_paths, args.stop_on_collapse)],\n", + " )\n", + " trainer.train()\n", + "\n", + " completions_dir = os.path.join(args.output_dir, \"completions\")\n", + " if os.environ.get(\"SM_OUTPUT_DATA_DIR\") and os.path.isdir(completions_dir):\n", + " shutil.copytree(\n", + " completions_dir,\n", + " os.path.join(os.environ[\"SM_OUTPUT_DATA_DIR\"], \"completions\"),\n", + " dirs_exist_ok=True,\n", + " )\n", + "\n", + " trainer.save_model(args.output_dir)\n", + " tokenizer.save_pretrained(args.output_dir)\n", + "\n", + "\n", + "if __name__ == \"__main__\":\n", + " main()\n" + ] + }, + { + "cell_type": "markdown", + "id": "b0249124", + "metadata": {}, + "source": [ + "ZeRO-2 shards optimizer state and gradients across the four L4 GPUs. That fits SmolLM3-3B with the batch settings below. For a larger model, switch to ZeRO-3 or lower `num_generations` / batch size.\n" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "16d3f677", + "metadata": {}, + "outputs": [], + "source": [ + "%%writefile src/launch.sh\n", + "#!/bin/sh\n", + "set -eu\n", + "\n", + "# Start one training process per GPU; SageMaker exposes the count as SM_NUM_GPUS.\n", + "unset NCCL_PROTO\n", + "exec torchrun --nnodes=1 --nproc_per_node=\"${SM_NUM_GPUS:-4}\" train.py \"$@\"\n" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "395b0169", + "metadata": {}, + "outputs": [], + "source": [ + "%%writefile src/ds_zero2.json\n", + "{\n", + " \"bf16\": { \"enabled\": \"auto\" },\n", + " \"zero_optimization\": {\n", + " \"stage\": 2,\n", + " \"overlap_comm\": true,\n", + " \"contiguous_gradients\": true,\n", + " \"reduce_bucket_size\": \"auto\"\n", + " },\n", + " \"gradient_accumulation_steps\": \"auto\",\n", + " \"gradient_clipping\": \"auto\",\n", + " \"train_batch_size\": \"auto\",\n", + " \"train_micro_batch_size_per_gpu\": \"auto\"\n", + "}" + ] + }, + { + "cell_type": "markdown", + "id": "6208d274", + "metadata": {}, + "source": [ + "## Launch the training job\n", + "\n", + "Multi-GPU GRPO needs one process per GPU. `launch.sh` is the SageMaker entry point because it can start `torchrun` with the GPU count SageMaker exposes:\n", + "\n", + "```sh\n", + "torchrun --nnodes=1 --nproc_per_node=\"$SM_NUM_GPUS\" train.py ...\n", + "```\n", + "\n", + "`launch.sh` only starts distributed execution; `train.py` contains the training logic.\n", + "\n", + "This configuration is intentionally small: it is meant to verify that the dataset, reward functions, distributed launch, and logging are wired correctly. Do not expect this short run to materially improve the model; for a real RL training run, increase the number of steps and validate on a held-out set.\n", + "\n", + "Set `DEBUG_NO_UPDATE=True` for a smoke test with `learning_rate=0.0` and `beta=0.0`. Leave it `False` to run a short training job.\n" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "3204defb", + "metadata": {}, + "outputs": [], + "source": [ + "import time\n", + "from sagemaker.train.model_trainer import ModelTrainer\n", + "from sagemaker.train.configs import SourceCode, Compute, InputData, OutputDataConfig, StoppingCondition\n", + "\n", + "DEBUG_NO_UPDATE = False\n", + "\n", + "base_job_name = \"smolm3-grpo-toolcall-\" + time.strftime(\"%Y%m%d-%H%M%S\", time.gmtime())\n", + "\n", + "trainer = ModelTrainer(\n", + " sagemaker_session=sess,\n", + " role=role,\n", + " training_image=TRAINING_IMAGE,\n", + " base_job_name=base_job_name,\n", + " source_code=SourceCode(source_dir=\"src\", entry_script=\"launch.sh\"),\n", + " compute=Compute(instance_type=INSTANCE_TYPE, instance_count=1, volume_size_in_gb=200),\n", + " stopping_condition=StoppingCondition(max_runtime_in_seconds=6 * 60 * 60),\n", + " output_data_config=OutputDataConfig(s3_output_path=f\"s3://{bucket}/xlam-grpo/output/\"),\n", + " environment={\n", + " \"HF_TOKEN\": HF_TOKEN,\n", + " \"NCCL_DEBUG\": \"WARN\",\n", + " },\n", + " hyperparameters={\n", + " \"model_id\": MODEL_ID,\n", + " \"max_steps\": 20 if DEBUG_NO_UPDATE else 50,\n", + " \"per_device_train_batch_size\": 1,\n", + " \"gradient_accumulation_steps\": 8, # global batch = 4 GPUs x 1 x 8\n", + " \"num_generations\": 8,\n", + " \"learning_rate\": 0.0 if DEBUG_NO_UPDATE else 1e-8,\n", + " \"beta\": 0.0 if DEBUG_NO_UPDATE else 0.1,\n", + " \"temperature\": 0.7,\n", + " \"top_p\": 1.0,\n", + " \"top_k\": 0,\n", + " \"max_grad_norm\": 0.05,\n", + " \"max_completion_length\": 128,\n", + " \"reward_weights\": \"1.0,0.5\", # exact match + lighter format shaping\n", + " \"repetition_penalty\": 1.05, # discourage max-length loops\n", + " \"log_completions\": True,\n", + " \"num_completions_to_print\": 8,\n", + " \"stop_on_collapse\": True,\n", + " \"remove_invalid_values\": False,\n", + " \"renormalize_logits\": True,\n", + " \"gradient_checkpointing\": True,\n", + " \"deepspeed\": \"ds_zero2.json\", # shard optimizer state across the 4 GPUs (required to fit)\n", + " },\n", + ")\n", + "\n", + "trainer.train(input_data_config=[InputData(channel_name=\"train\", data_source=train_s3)], wait=True)" + ] + }, + { + "cell_type": "markdown", + "id": "6aba923c", + "metadata": {}, + "source": [ + "## Training curves\n", + "\n", + "`GRPOTrainer` logs reward metrics during training. Plot those first: they show whether the model is still producing parseable tool calls and whether exact-match reward stays alive.\n", + "\n", + "A noisy reward curve is normal. A fast drop to zero with clipped completions means generation collapsed.\n" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "57276d42", + "metadata": {}, + "outputs": [], + "source": [ + "import boto3\n", + "\n", + "TRAINING_JOB_NAME = \"\" # set to a specific completed job name, or leave blank for latest\n", + "JOB_PREFIX = \"smolm3-grpo-toolcall-\"\n", + "\n", + "sm = boto3.client(\"sagemaker\", region_name=REGION)\n", + "\n", + "if TRAINING_JOB_NAME:\n", + " job = TRAINING_JOB_NAME\n", + "elif \"trainer\" in globals() and hasattr(trainer, \"latest_training_job\"):\n", + " job = trainer.latest_training_job.name\n", + "elif \"base_job_name\" in globals():\n", + " job = base_job_name\n", + "else:\n", + " jobs = sm.list_training_jobs(\n", + " NameContains=JOB_PREFIX,\n", + " SortBy=\"CreationTime\",\n", + " SortOrder=\"Descending\",\n", + " MaxResults=1,\n", + " )[\"TrainingJobSummaries\"]\n", + " assert jobs, f\"No SageMaker jobs found with prefix {JOB_PREFIX!r}; set TRAINING_JOB_NAME manually.\"\n", + " job = jobs[0][\"TrainingJobName\"]\n", + "\n", + "desc = sm.describe_training_job(TrainingJobName=job)\n", + "print(\"job: \", job)\n", + "print(\"status:\", desc[\"TrainingJobStatus\"])\n", + "print(\"model: \", desc.get(\"ModelArtifacts\", {}).get(\"S3ModelArtifacts\"))\n", + "print(f\"console: https://{REGION}.console.aws.amazon.com/sagemaker/home?region={REGION}#/jobs/{job}\")\n" + ] + }, + { + "cell_type": "markdown", + "id": "1d17afaa", + "metadata": {}, + "source": [ + "### Plot reward metrics\n", + "\n", + "Read metrics from the SageMaker output artifact. If the artifact is not ready, use CloudWatch logs or notebook output.\n" + ] + }, + { + "cell_type": "code", + "execution_count": null, + "id": "ffaf61a9", + "metadata": {}, + "outputs": [], + "source": [ + "import ast\n", + "import html\n", + "import json\n", + "import os\n", + "import re\n", + "import tarfile\n", + "import tempfile\n", + "from pathlib import Path\n", + "from urllib.parse import urlparse\n", + "\n", + "from botocore.exceptions import ClientError\n", + "\n", + "os.environ.setdefault(\"MPLCONFIGDIR\", \"/tmp/matplotlib\")\n", + "import matplotlib.pyplot as plt\n", + "import pandas as pd\n", + "\n", + "metric_re = re.compile(r\"\\{.*?(?:\\'reward\\'|\\\"reward\\\").*?\\}\", re.DOTALL)\n", + "ansi_re = re.compile(r\"\\x1b\\[[0-?]*[ -/]*[@-~]\")\n", + "tag_re = re.compile(r\"<[^>]+>\")\n", + "\n", + "\n", + "def maybe_float(value):\n", + " try:\n", + " return float(value)\n", + " except (TypeError, ValueError):\n", + " return value\n", + "\n", + "\n", + "def parse_metric_dicts(chunks):\n", + " records, seen = [], set()\n", + " for chunk in chunks:\n", + " text = html.unescape(str(chunk))\n", + " text = tag_re.sub(\"\", text)\n", + " text = ansi_re.sub(\"\", text).replace(\"#015\", \"\").replace(\"\\r\", \"\")\n", + " for match in metric_re.finditer(text):\n", + " raw = \" \".join(match.group(0).split())\n", + " if raw in seen:\n", + " continue\n", + " seen.add(raw)\n", + " try:\n", + " record = ast.literal_eval(raw)\n", + " except (SyntaxError, ValueError):\n", + " continue\n", + " if isinstance(record, dict) and \"reward\" in record:\n", + " records.append({k: maybe_float(v) for k, v in record.items()})\n", + " return records\n", + "\n", + "\n", + "def s3_metric_records(training_desc):\n", + " artifact = training_desc.get(\"ModelArtifacts\", {}).get(\"S3ModelArtifacts\")\n", + " assert artifact, \"No model artifact URI found.\"\n", + " output_uri = artifact.rsplit(\"/\", 1)[0] + \"/output.tar.gz\"\n", + " parsed = urlparse(output_uri)\n", + " local_tar = Path(tempfile.gettempdir()) / f\"{job}-output.tar.gz\"\n", + " boto3.client(\"s3\", region_name=REGION).download_file(parsed.netloc, parsed.path.lstrip(\"/\"), str(local_tar))\n", + " with tarfile.open(local_tar) as tar:\n", + " member = next(m for m in tar.getmembers() if m.name.endswith(\"training_metrics.jsonl\"))\n", + " lines = tar.extractfile(member).read().decode().splitlines()\n", + " Path(\"training_metrics.jsonl\").write_text(\"\\n\".join(lines) + \"\\n\")\n", + " return [json.loads(line) for line in lines if line.strip()]\n", + "\n", + "\n", + "def cloudwatch_metric_records(job_name):\n", + " logs = boto3.client(\"logs\", region_name=REGION)\n", + " kwargs = {\n", + " \"logGroupName\": \"/aws/sagemaker/TrainingJobs\",\n", + " \"logStreamNamePrefix\": job_name,\n", + " \"startFromHead\": True,\n", + " }\n", + " chunks = []\n", + " while True:\n", + " page = logs.filter_log_events(**kwargs)\n", + " chunks.extend(event.get(\"message\", \"\") for event in page.get(\"events\", []))\n", + " token = page.get(\"nextToken\")\n", + " if not token:\n", + " break\n", + " kwargs[\"nextToken\"] = token\n", + " return parse_metric_dicts(chunks)\n", + "\n", + "\n", + "def notebook_metric_records(path=\"notebook.ipynb\"):\n", + " nb = json.loads(Path(path).read_text())\n", + " chunks = []\n", + " for cell in nb[\"cells\"]:\n", + " for output in cell.get(\"outputs\", []):\n", + " if \"text\" in output:\n", + " value = output[\"text\"]\n", + " chunks.append(\"\".join(value) if isinstance(value, list) else value)\n", + " for value in output.get(\"data\", {}).values():\n", + " chunks.append(\"\".join(value) if isinstance(value, list) else str(value))\n", + " return parse_metric_dicts(chunks)\n", + "\n", + "\n", + "try:\n", + " records = s3_metric_records(desc)\n", + "except Exception as e:\n", + " print(f\"Could not read metrics from S3 output artifact: {e}\")\n", + " metrics_file = Path(\"training_metrics.jsonl\")\n", + " if metrics_file.exists():\n", + " records = [json.loads(line) for line in metrics_file.read_text().splitlines() if line.strip()]\n", + " else:\n", + " try:\n", + " records = cloudwatch_metric_records(job)\n", + " except ClientError as e:\n", + " if e.response.get(\"Error\", {}).get(\"Code\") != \"AccessDeniedException\":\n", + " raise\n", + " print(\"CloudWatch Logs access denied; parsing saved notebook output instead.\")\n", + " records = notebook_metric_records()\n", + "\n", + "records = [r for r in records if \"reward\" in r]\n", + "assert records, f\"No GRPO reward metrics found for {job}.\"\n", + "\n", + "metrics = pd.DataFrame(records)\n", + "metrics[\"step\"] = range(1, len(metrics) + 1)\n", + "metrics[\"reward_smooth\"] = metrics[\"reward\"].rolling(5, min_periods=1).mean()\n", + "metrics.to_csv(\"training_metrics.csv\", index=False)\n", + "\n", + "cols = [\n", + " \"step\",\n", + " \"reward\",\n", + " \"reward_smooth\",\n", + " \"rewards/tool_call_reward/mean\",\n", + " \"rewards/format_reward/mean\",\n", + " \"completions/clipped_ratio\",\n", + " \"loss\",\n", + " \"entropy\",\n", + "]\n", + "display(metrics[[c for c in cols if c in metrics]].head())\n", + "\n", + "plot_cols = [\n", + " \"reward\",\n", + " \"reward_smooth\",\n", + " \"rewards/tool_call_reward/mean\",\n", + " \"rewards/format_reward/mean\",\n", + "]\n", + "plot_cols = [c for c in plot_cols if c in metrics]\n", + "\n", + "ax = metrics.plot(x=\"step\", y=plot_cols, figsize=(10, 5), marker=\"o\")\n", + "ax.set_title(f\"GRPO reward curves: {MODEL_ID}\")\n", + "ax.set_xlabel(\"logged training step\")\n", + "ax.set_ylabel(\"reward\")\n", + "ax.grid(True, alpha=0.3)\n", + "plt.tight_layout()\n", + "plt.savefig(\"training_reward_curves.png\", dpi=160)\n", + "plt.show()\n" + ] + }, + { + "cell_type": "markdown", + "metadata": {}, + "source": [ + "## Conclusion\n", + "\n", + "You now have the full shape of a verifiable-reward RL training job on SageMaker: a dataset with hidden ground-truth tool calls, reward functions that score generated calls, a TRL GRPO training script, and a multi-GPU SageMaker launch.\n", + "\n", + "The run above is deliberately short so the notebook stays practical to execute. To turn it into a real training run, train for more steps, use a held-out evaluation split, and track exact-match tool-call accuracy in addition to the reward curves.\n" + ] + } + ], + "metadata": { + "kernelspec": { + "display_name": "Python 3 (ipykernel)", + "language": "python", + "name": "python3" + }, + "language_info": { + "codemirror_mode": { + "name": "ipython", + "version": 3 + }, + "file_extension": ".py", + "mimetype": "text/x-python", + "name": "python", + "nbconvert_exporter": "python", + "pygments_lexer": "ipython3", + "version": "3.14.5" + } + }, + "nbformat": 4, + "nbformat_minor": 5 +} From 5afe9b4d66c8b5743ade00ec4e54ce20afea7c37 Mon Sep 17 00:00:00 2001 From: DWarez Date: Fri, 26 Jun 2026 10:44:33 +0200 Subject: [PATCH 2/5] change: using ml.p4d.24xlarge instance to enable beta>0.0 change: using zero3 instead of zero2 Signed-off-by: DWarez --- .../grpo-llm-trl/sagemaker-notebook.ipynb | 5889 ++++++++++++++++- 1 file changed, 5868 insertions(+), 21 deletions(-) diff --git a/docs/sagemaker/notebooks/sagemaker-sdk/grpo-llm-trl/sagemaker-notebook.ipynb b/docs/sagemaker/notebooks/sagemaker-sdk/grpo-llm-trl/sagemaker-notebook.ipynb index bf35d8e06..d6cf4c6da 100644 --- a/docs/sagemaker/notebooks/sagemaker-sdk/grpo-llm-trl/sagemaker-notebook.ipynb +++ b/docs/sagemaker/notebooks/sagemaker-sdk/grpo-llm-trl/sagemaker-notebook.ipynb @@ -20,7 +20,7 @@ "- launch a multi-GPU training job,\n", "- plot the reward curves from the training logs.\n", "\n", - "You need AWS credentials, a SageMaker execution role with S3 access, quota for `ml.g6.12xlarge` training in `us-east-1`, and a Hugging Face token that can download the base model.\n" + "You need AWS credentials, a SageMaker execution role with S3 access, quota for `ml.p4d.24xlarge` training in `us-east-1`, and a Hugging Face token that can download the base model.\n" ] }, { @@ -121,7 +121,7 @@ "outputs": [], "source": [ "MODEL_ID = \"HuggingFaceTB/SmolLM3-3B\" # SmolLM3, HF's 3B instruct model\n", - "INSTANCE_TYPE = \"ml.g6.12xlarge\" # 4xL4 24GB\n", + "INSTANCE_TYPE = \"ml.p4d.24xlarge\" # 8xA100 40GB\n", "\n", "TRAINING_IMAGE = \"754289655784.dkr.ecr.us-east-1.amazonaws.com/hf-trl-grpo:sagemaker-trl-dev-e63f67e\"" ] @@ -454,10 +454,18 @@ }, { "cell_type": "code", - "execution_count": null, + "execution_count": 24, "id": "f42a218b", "metadata": {}, - "outputs": [], + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Overwriting src/train.py\n" + ] + } + ], "source": [ "%%writefile src/train.py\n", "'''SageMaker entry script: GRPO tool-call training with TRL.\n", @@ -530,7 +538,7 @@ " p.add_argument(\"--gradient_checkpointing\", type=str2bool, default=True)\n", " p.add_argument(\"--deepspeed\", default=\"\")\n", " p.add_argument(\"--report_to\", default=\"none\")\n", - " p.add_argument(\"--log_completions\", type=str2bool, default=True)\n", + " p.add_argument(\"--log_completions\", type=str2bool, default=False)\n", " p.add_argument(\"--num_completions_to_print\", type=int, default=8)\n", " p.add_argument(\"--stop_on_collapse\", type=str2bool, default=True)\n", " p.add_argument(\"--remove_invalid_values\", type=str2bool, default=False)\n", @@ -624,22 +632,30 @@ "id": "b0249124", "metadata": {}, "source": [ - "ZeRO-2 shards optimizer state and gradients across the four L4 GPUs. That fits SmolLM3-3B with the batch settings below. For a larger model, switch to ZeRO-3 or lower `num_generations` / batch size.\n" + "ZeRO-3 shards the model, gradients, and optimizer state across the eight A100 GPUs. This run keeps `beta` enabled, so TRL also loads a reference model for the KL term.\n" ] }, { "cell_type": "code", - "execution_count": null, + "execution_count": 25, "id": "16d3f677", "metadata": {}, - "outputs": [], + "outputs": [ + { + "name": "stdout", + "output_type": "stream", + "text": [ + "Overwriting src/launch.sh\n" + ] + } + ], "source": [ "%%writefile src/launch.sh\n", "#!/bin/sh\n", "set -eu\n", "\n", "# Start one training process per GPU; SageMaker exposes the count as SM_NUM_GPUS.\n", - "unset NCCL_PROTO\n", + "# unset NCCL_PROTO\n", "exec torchrun --nnodes=1 --nproc_per_node=\"${SM_NUM_GPUS:-4}\" train.py \"$@\"\n" ] }, @@ -650,20 +666,23 @@ "metadata": {}, "outputs": [], "source": [ - "%%writefile src/ds_zero2.json\n", + "%%writefile src/ds_zero3.json\n", "{\n", " \"bf16\": { \"enabled\": \"auto\" },\n", " \"zero_optimization\": {\n", - " \"stage\": 2,\n", + " \"stage\": 3,\n", " \"overlap_comm\": true,\n", " \"contiguous_gradients\": true,\n", - " \"reduce_bucket_size\": \"auto\"\n", + " \"reduce_bucket_size\": \"auto\",\n", + " \"stage3_prefetch_bucket_size\": \"auto\",\n", + " \"stage3_param_persistence_threshold\": \"auto\",\n", + " \"stage3_gather_16bit_weights_on_model_save\": true\n", " },\n", " \"gradient_accumulation_steps\": \"auto\",\n", " \"gradient_clipping\": \"auto\",\n", " \"train_batch_size\": \"auto\",\n", " \"train_micro_batch_size_per_gpu\": \"auto\"\n", - "}" + "}\n" ] }, { @@ -683,15 +702,5842 @@ "\n", "This configuration is intentionally small: it is meant to verify that the dataset, reward functions, distributed launch, and logging are wired correctly. Do not expect this short run to materially improve the model; for a real RL training run, increase the number of steps and validate on a held-out set.\n", "\n", - "Set `DEBUG_NO_UPDATE=True` for a smoke test with `learning_rate=0.0` and `beta=0.0`. Leave it `False` to run a short training job.\n" + "Set `DEBUG_NO_UPDATE=True` for a smoke test with `learning_rate=0.0`. `beta` stays enabled so the run still exercises the reference model and KL path.\n" ] }, { "cell_type": "code", - "execution_count": null, + "execution_count": 26, "id": "3204defb", "metadata": {}, - "outputs": [], + "outputs": [ + { + "data": { + "text/html": [ + "
[06/26/26 10:18:12] INFO     OutputDataConfig compression type not provided. Using default:         defaults.py:165\n",
+       "                             GZIP                                                                                  \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:18:12]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m OutputDataConfig compression type not provided. Using default: \u001b]8;id=17026267;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/train/defaults.py\u001b\\\u001b[2mdefaults.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026268;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/train/defaults.py#165\u001b\\\u001b[2m165\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m GZIP \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     Training image URI:                                               model_trainer.py:558\n",
+       "                             754289655784.dkr.ecr.us-east-1.amazonaws.com/hf-trl-grpo:sagemake                     \n",
+       "                             r-trl-dev-e63f67e                                                                     \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m Training image URI: \u001b]8;id=17026275;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/train/model_trainer.py\u001b\\\u001b[2mmodel_trainer.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026276;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/train/model_trainer.py#558\u001b\\\u001b[2m558\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m754289655784.\u001b[0mdkr.ecr.us-east-\u001b[1m1.\u001b[0mamazonaws.com/hf-trl-grpo:sagemake \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m r-trl-dev-e63f67e \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     SageMaker Python SDK will collect telemetry to help us better telemetry_logging.py:110\n",
+       "                             understand our user's needs, diagnose issues, and deliver                             \n",
+       "                             additional features.                                                                  \n",
+       "                             To opt out of telemetry, please disable via TelemetryOptOut                           \n",
+       "                             parameter in SDK defaults config. For more information, refer                         \n",
+       "                             to                                                                                    \n",
+       "                             https://sagemaker.readthedocs.io/en/stable/overview.html#conf                         \n",
+       "                             iguring-and-using-defaults-with-the-sagemaker-python-sdk.                             \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m SageMaker Python SDK will collect telemetry to help us better \u001b]8;id=17026283;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/telemetry/telemetry_logging.py\u001b\\\u001b[2mtelemetry_logging.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026284;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/telemetry/telemetry_logging.py#110\u001b\\\u001b[2m110\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m understand our user's needs, diagnose issues, and deliver \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m additional features. \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m To opt out of telemetry, please disable via TelemetryOptOut \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m parameter in SDK defaults config. For more information, refer \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m to \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[4mhttps://sagemaker.readthedocs.io/en/stable/overview.html#conf\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[4miguring-and-using-defaults-with-the-sagemaker-python-sdk.\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:18:19] INFO     Creating training_job resource.                                     resources.py:31116\n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:18:19]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m Creating training_job resource. \u001b]8;id=17026291;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026292;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31116\u001b\\\u001b[2m31116\u001b[0m\u001b]8;;\u001b\\\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
/Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/rich/live.py:260: UserWarning: \n",
+       "install \"ipywidgets\" for Jupyter support\n",
+       "  warnings.warn('install \"ipywidgets\" for Jupyter support')\n",
+       "
\n" + ], + "text/plain": [ + "/Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/rich/live.py:260: UserWarning: \n", + "install \"ipywidgets\" for Jupyter support\n", + " warnings.warn('install \"ipywidgets\" for Jupyter support')\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:29:25] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             Starting training script                                                              \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:29:25]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026299;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026300;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Starting training script \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             ++ /opt/venv/bin/python3 --version                                                    \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026307;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026308;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m ++ /opt/venv/bin/python3 --version \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             Python 3.12.13                                                                        \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026315;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026316;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Python \u001b[1m3.12\u001b[0m.\u001b[1m13\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             ++ echo /opt/ml/input/config/resourceconfig.json:                                     \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026323;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026324;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m ++ echo /opt/ml/input/config/resourceconfig.json: \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             ++ cat /opt/ml/input/config/resourceconfig.json                                       \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026331;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026332;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m ++ cat /opt/ml/input/config/resourceconfig.json \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             /opt/ml/input/config/resourceconfig.json:                                             \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026339;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026340;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m /opt/ml/input/config/resourceconfig.json: \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             ++ echo                                                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026347;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026348;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m ++ echo \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             {\"current_host\":\"algo-1\",\"current_instance_type\":\"ml.p4d.24xlarge\",                   \n",
+       "                             \"current_group_name\":\"homogeneousCluster\",\"hosts\":[\"algo-1\"],\"insta                   \n",
+       "                             nce_groups\":[{\"instance_group_name\":\"homogeneousCluster\",\"instance_                   \n",
+       "                             type\":\"ml.p4d.24xlarge\",\"hosts\":[\"algo-1\"]}],\"network_interface_nam                   \n",
+       "                             e\":\"eth0\",\"topology\":null}                                                            \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026355;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026356;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m{\u001b[0m\"current_host\":\"algo-1\",\"current_instance_type\":\"ml.p4d.24xlarge\", \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"current_group_name\":\"homogeneousCluster\",\"hosts\":\u001b[1m[\u001b[0m\"algo-1\"\u001b[1m]\u001b[0m,\"insta \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m nce_groups\":\u001b[1m[\u001b[0m\u001b[1m{\u001b[0m\"instance_group_name\":\"homogeneousCluster\",\"instance_ \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m type\":\"ml.p4d.24xlarge\",\"hosts\":\u001b[1m[\u001b[0m\"algo-1\"\u001b[1m]\u001b[0m\u001b[1m}\u001b[0m\u001b[1m]\u001b[0m,\"network_interface_nam \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m e\":\"eth0\",\"topology\":null\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             ++ echo /opt/ml/input/config/inputdataconfig.json:                                    \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026363;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026364;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m ++ echo /opt/ml/input/config/inputdataconfig.json: \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             /opt/ml/input/config/inputdataconfig.json:                                            \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026371;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026372;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m /opt/ml/input/config/inputdataconfig.json: \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             ++ cat /opt/ml/input/config/inputdataconfig.json                                      \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026379;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026380;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m ++ cat /opt/ml/input/config/inputdataconfig.json \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             ++ echo                                                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026387;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026388;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m ++ echo \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             ++ echo 'Setting up environment variables'                                            \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026395;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026396;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m ++ echo 'Setting up environment variables' \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             {\"code\":{\"TrainingInputMode\":\"File\",\"S3DistributionType\":\"FullyRepl                   \n",
+       "                             icated\",\"RecordWrapperType\":\"None\"},\"sm_drivers\":{\"TrainingInputMod                   \n",
+       "                             e\":\"File\",\"S3DistributionType\":\"FullyReplicated\",\"RecordWrapperType                   \n",
+       "                             \":\"None\"},\"train\":{\"TrainingInputMode\":\"File\",\"S3DistributionType\":                   \n",
+       "                             \"FullyReplicated\",\"RecordWrapperType\":\"None\"}}                                        \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026403;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026404;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m{\u001b[0m\"code\":\u001b[1m{\u001b[0m\"TrainingInputMode\":\"File\",\"S3DistributionType\":\"FullyRepl \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m icated\",\"RecordWrapperType\":\"None\"\u001b[1m}\u001b[0m,\"sm_drivers\":\u001b[1m{\u001b[0m\"TrainingInputMod \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m e\":\"File\",\"S3DistributionType\":\"FullyReplicated\",\"RecordWrapperType \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \":\"None\"\u001b[1m}\u001b[0m,\"train\":\u001b[1m{\u001b[0m\"TrainingInputMode\":\"File\",\"S3DistributionType\": \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"FullyReplicated\",\"RecordWrapperType\":\"None\"\u001b[1m}\u001b[0m\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             ++ /opt/venv/bin/python3                                                              \n",
+       "                             /opt/ml/input/data/sm_drivers/scripts/environment.py                                  \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026411;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026412;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m ++ /opt/venv/bin/python3 \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m /opt/ml/input/data/sm_drivers/scripts/environment.py \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             Setting up environment variables                                                      \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026419;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026420;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Setting up environment variables \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             No Neurons detected (normal if no neurons installed)                                  \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026427;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026428;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m No Neurons detected \u001b[1m(\u001b[0mnormal if no neurons installed\u001b[1m)\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             Environment Variables:                                                                \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026435;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026436;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Environment Variables: \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             NV_LIBCUBLAS_VERSION=13.1.0.3-1                                                       \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026443;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026444;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m NV_LIBCUBLAS_VERSION=13.1.0.3-\u001b[1m1\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             NVIDIA_VISIBLE_DEVICES=all                                                            \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026451;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026452;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m NVIDIA_VISIBLE_DEVICES=all \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             PYTHONUNBUFFERED=1                                                                    \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026459;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026460;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m PYTHONUNBUFFERED=\u001b[1m1\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             AWS_CONTAINER_CREDENTIALS_RELATIVE_URI=******                                         \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026467;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026468;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m AWS_CONTAINER_CREDENTIALS_RELATIVE_URI=****** \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SAGEMAKER_TRAINING_MODULE=sagemaker_pytorch_container.training:main                   \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026475;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026476;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SAGEMAKER_TRAINING_MODULE=sagemaker_pytorch_container.training:main \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             HOSTNAME=ip-10-0-196-59.ec2.internal                                                  \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026483;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026484;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m HOSTNAME=ip-\u001b[1m10\u001b[0m-\u001b[1m0\u001b[0m-\u001b[1m196\u001b[0m-\u001b[1m59.\u001b[0mec2.internal \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             NVIDIA_REQUIRE_CUDA=cuda>=13.0 brand=unknown,driver>=535,driver<536                   \n",
+       "                             brand=grid,driver>=535,driver<536                                                     \n",
+       "                             brand=tesla,driver>=535,driver<536                                                    \n",
+       "                             brand=nvidia,driver>=535,driver<536                                                   \n",
+       "                             brand=quadro,driver>=535,driver<536                                                   \n",
+       "                             brand=quadrortx,driver>=535,driver<536                                                \n",
+       "                             brand=nvidiartx,driver>=535,driver<536                                                \n",
+       "                             brand=vapps,driver>=535,driver<536 brand=vpc,driver>=535,driver<536                   \n",
+       "                             brand=vcs,driver>=535,driver<536 brand=vws,driver>=535,driver<536                     \n",
+       "                             brand=cloudgaming,driver>=535,driver<536                                              \n",
+       "                             brand=unknown,driver>=550,driver<551                                                  \n",
+       "                             brand=grid,driver>=550,driver<551                                                     \n",
+       "                             brand=tesla,driver>=550,driver<551                                                    \n",
+       "                             brand=nvidia,driver>=550,driver<551                                                   \n",
+       "                             brand=quadro,driver>=550,driver<551                                                   \n",
+       "                             brand=quadrortx,driver>=550,driver<551                                                \n",
+       "                             brand=nvidiartx,driver>=550,driver<551                                                \n",
+       "                             brand=vapps,driver>=550,driver<551 brand=vpc,driver>=550,driver<551                   \n",
+       "                             brand=vcs,driver>=550,driver<551 brand=vws,driver>=550,driver<551                     \n",
+       "                             brand=cloudgaming,driver>=550,driver<551                                              \n",
+       "                             brand=unknown,driver>=565,driver<566                                                  \n",
+       "                             brand=grid,driver>=565,driver<566                                                     \n",
+       "                             brand=tesla,driver>=565,driver<566                                                    \n",
+       "                             brand=nvidia,driver>=565,driver<566                                                   \n",
+       "                             brand=quadro,driver>=565,driver<566                                                   \n",
+       "                             brand=quadrortx,driver>=565,driver<566                                                \n",
+       "                             brand=nvidiartx,driver>=565,driver<566                                                \n",
+       "                             brand=vapps,driver>=565,driver<566 brand=vpc,driver>=565,driver<566                   \n",
+       "                             brand=vcs,driver>=565,driver<566 brand=vws,driver>=565,driver<566                     \n",
+       "                             brand=cloudgaming,driver>=565,driver<566                                              \n",
+       "                             brand=unknown,driver>=570,driver<571                                                  \n",
+       "                             brand=grid,driver>=570,driver<571                                                     \n",
+       "                             brand=tesla,driver>=570,driver<571                                                    \n",
+       "                             brand=nvidia,driver>=570,driver<571                                                   \n",
+       "                             brand=quadro,driver>=570,driver<571                                                   \n",
+       "                             brand=quadrortx,driver>=570,driver<571                                                \n",
+       "                             brand=nvidiartx,driver>=570,driver<571                                                \n",
+       "                             brand=vapps,driver>=570,driver<571 brand=vpc,driver>=570,driver<571                   \n",
+       "                             brand=vcs,driver>=570,driver<571 brand=vws,driver>=570,driver<571                     \n",
+       "                             brand=cloudgaming,driver>=570,driver<571                                              \n",
+       "                             brand=unknown,driver>=575,driver<576                                                  \n",
+       "                             brand=grid,driver>=575,driver<576                                                     \n",
+       "                             brand=tesla,driver>=575,driver<576                                                    \n",
+       "                             brand=nvidia,driver>=575,driver<576                                                   \n",
+       "                             brand=quadro,driver>=575,driver<576                                                   \n",
+       "                             brand=quadrortx,driver>=575,driver<576                                                \n",
+       "                             brand=nvidiartx,driver>=575,driver<576                                                \n",
+       "                             brand=vapps,driver>=575,driver<576 brand=vpc,driver>=575,driver<576                   \n",
+       "                             brand=vcs,driver>=575,driver<576 brand=vws,driver>=575,driver<576                     \n",
+       "                             brand=cloudgaming,driver>=575,driver<576                                              \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026491;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026492;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m NVIDIA_REQUIRE_CUDA=cuda>=\u001b[1m13.0\u001b[0m brand=unknown,driver>=\u001b[1m535\u001b[0m,driver\u001b[1m<\u001b[0m\u001b[1m536\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=grid,driver>=\u001b[1m535\u001b[0m,driver<\u001b[1m536\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=tesla,driver>=\u001b[1m535\u001b[0m,driver<\u001b[1m536\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=nvidia,driver>=\u001b[1m535\u001b[0m,driver<\u001b[1m536\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=quadro,driver>=\u001b[1m535\u001b[0m,driver<\u001b[1m536\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=quadrortx,driver>=\u001b[1m535\u001b[0m,driver<\u001b[1m536\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=nvidiartx,driver>=\u001b[1m535\u001b[0m,driver<\u001b[1m536\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=vapps,driver>=\u001b[1m535\u001b[0m,driver<\u001b[1m536\u001b[0m brand=vpc,driver>=\u001b[1m535\u001b[0m,driver<\u001b[1m536\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=vcs,driver>=\u001b[1m535\u001b[0m,driver<\u001b[1m536\u001b[0m brand=vws,driver>=\u001b[1m535\u001b[0m,driver<\u001b[1m536\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=cloudgaming,driver>=\u001b[1m535\u001b[0m,driver<\u001b[1m536\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=unknown,driver>=\u001b[1m550\u001b[0m,driver<\u001b[1m551\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=grid,driver>=\u001b[1m550\u001b[0m,driver<\u001b[1m551\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=tesla,driver>=\u001b[1m550\u001b[0m,driver<\u001b[1m551\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=nvidia,driver>=\u001b[1m550\u001b[0m,driver<\u001b[1m551\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=quadro,driver>=\u001b[1m550\u001b[0m,driver<\u001b[1m551\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=quadrortx,driver>=\u001b[1m550\u001b[0m,driver<\u001b[1m551\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=nvidiartx,driver>=\u001b[1m550\u001b[0m,driver<\u001b[1m551\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=vapps,driver>=\u001b[1m550\u001b[0m,driver<\u001b[1m551\u001b[0m brand=vpc,driver>=\u001b[1m550\u001b[0m,driver<\u001b[1m551\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=vcs,driver>=\u001b[1m550\u001b[0m,driver<\u001b[1m551\u001b[0m brand=vws,driver>=\u001b[1m550\u001b[0m,driver<\u001b[1m551\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=cloudgaming,driver>=\u001b[1m550\u001b[0m,driver<\u001b[1m551\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=unknown,driver>=\u001b[1m565\u001b[0m,driver<\u001b[1m566\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=grid,driver>=\u001b[1m565\u001b[0m,driver<\u001b[1m566\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=tesla,driver>=\u001b[1m565\u001b[0m,driver<\u001b[1m566\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=nvidia,driver>=\u001b[1m565\u001b[0m,driver<\u001b[1m566\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=quadro,driver>=\u001b[1m565\u001b[0m,driver<\u001b[1m566\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=quadrortx,driver>=\u001b[1m565\u001b[0m,driver<\u001b[1m566\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=nvidiartx,driver>=\u001b[1m565\u001b[0m,driver<\u001b[1m566\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=vapps,driver>=\u001b[1m565\u001b[0m,driver<\u001b[1m566\u001b[0m brand=vpc,driver>=\u001b[1m565\u001b[0m,driver<\u001b[1m566\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=vcs,driver>=\u001b[1m565\u001b[0m,driver<\u001b[1m566\u001b[0m brand=vws,driver>=\u001b[1m565\u001b[0m,driver<\u001b[1m566\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=cloudgaming,driver>=\u001b[1m565\u001b[0m,driver<\u001b[1m566\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=unknown,driver>=\u001b[1m570\u001b[0m,driver<\u001b[1m571\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=grid,driver>=\u001b[1m570\u001b[0m,driver<\u001b[1m571\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=tesla,driver>=\u001b[1m570\u001b[0m,driver<\u001b[1m571\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=nvidia,driver>=\u001b[1m570\u001b[0m,driver<\u001b[1m571\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=quadro,driver>=\u001b[1m570\u001b[0m,driver<\u001b[1m571\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=quadrortx,driver>=\u001b[1m570\u001b[0m,driver<\u001b[1m571\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=nvidiartx,driver>=\u001b[1m570\u001b[0m,driver<\u001b[1m571\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=vapps,driver>=\u001b[1m570\u001b[0m,driver<\u001b[1m571\u001b[0m brand=vpc,driver>=\u001b[1m570\u001b[0m,driver<\u001b[1m571\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=vcs,driver>=\u001b[1m570\u001b[0m,driver<\u001b[1m571\u001b[0m brand=vws,driver>=\u001b[1m570\u001b[0m,driver<\u001b[1m571\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=cloudgaming,driver>=\u001b[1m570\u001b[0m,driver<\u001b[1m571\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=unknown,driver>=\u001b[1m575\u001b[0m,driver<\u001b[1m576\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=grid,driver>=\u001b[1m575\u001b[0m,driver<\u001b[1m576\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=tesla,driver>=\u001b[1m575\u001b[0m,driver<\u001b[1m576\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=nvidia,driver>=\u001b[1m575\u001b[0m,driver<\u001b[1m576\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=quadro,driver>=\u001b[1m575\u001b[0m,driver<\u001b[1m576\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=quadrortx,driver>=\u001b[1m575\u001b[0m,driver<\u001b[1m576\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=nvidiartx,driver>=\u001b[1m575\u001b[0m,driver<\u001b[1m576\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=vapps,driver>=\u001b[1m575\u001b[0m,driver<\u001b[1m576\u001b[0m brand=vpc,driver>=\u001b[1m575\u001b[0m,driver<\u001b[1m576\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=vcs,driver>=\u001b[1m575\u001b[0m,driver<\u001b[1m576\u001b[0m brand=vws,driver>=\u001b[1m575\u001b[0m,driver<\u001b[1m576\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m brand=cloudgaming,driver\u001b[1m>\u001b[0m=\u001b[1m575\u001b[0m,driver<\u001b[1m576\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             NV_NVTX_VERSION=13.0.85-1                                                             \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026499;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026500;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m NV_NVTX_VERSION=\u001b[1m13\u001b[0m\u001b[1m.0\u001b[0m.\u001b[1m85\u001b[0m-\u001b[1m1\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             NV_LIBNPP_VERSION=13.0.1.2-1                                                          \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026507;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026508;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m NV_LIBNPP_VERSION=13.0.1.2-\u001b[1m1\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             AWS_REGION=us-east-1                                                                  \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026515;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026516;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m AWS_REGION=us-east-\u001b[1m1\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             PWD=/workspace                                                                        \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026523;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026524;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m PWD=/workspace \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SAGEMAKER_MANAGED_WARMPOOL_CACHE_DIRECTORY=/opt/ml/sagemaker/warmpo                   \n",
+       "                             olcache                                                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026531;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026532;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SAGEMAKER_MANAGED_WARMPOOL_CACHE_DIRECTORY=/opt/ml/sagemaker/warmpo \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m olcache \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             NVIDIA_DRIVER_CAPABILITIES=compute,utility,compat32,graphics,video                    \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026539;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026540;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m NVIDIA_DRIVER_CAPABILITIES=compute,utility,compat32,graphics,video \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             NV_LIBNPP_PACKAGE=libnpp-13-0-13.0.1.2-1                                              \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026547;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026548;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m NV_LIBNPP_PACKAGE=libnpp-\u001b[1m13\u001b[0m-\u001b[1m0\u001b[0m-13.0.1.2-\u001b[1m1\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             NCCL_DEBUG=WARN                                                                       \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026555;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026556;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m NCCL_DEBUG=WARN \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             NVIDIA_PRODUCT_NAME=CUDA                                                              \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026563;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026564;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m NVIDIA_PRODUCT_NAME=CUDA \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             NV_CUDA_CUDART_VERSION=13.0.96-1                                                      \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026571;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026572;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m NV_CUDA_CUDART_VERSION=\u001b[1m13\u001b[0m\u001b[1m.0\u001b[0m.\u001b[1m96\u001b[0m-\u001b[1m1\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             HOME=/root                                                                            \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026579;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026580;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m HOME=/root \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             LANG=C.UTF-8                                                                          \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026587;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026588;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m LANG=C.UTF-\u001b[1m8\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             CUDA_VERSION=13.0.2                                                                   \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026595;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026596;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m CUDA_VERSION=\u001b[1m13\u001b[0m\u001b[1m.0\u001b[0m.\u001b[1m2\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             HF_HUB_USER_AGENT_ORIGIN=aws:sagemaker:gpu-cuda:training:hf-pytorch                   \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026603;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026604;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m HF_HUB_USER_AGENT_ORIGIN=aws:sagemaker:gpu-cuda:training:hf-pytorch \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             DMLC_INTERFACE=eth0                                                                   \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026611;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026612;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m DMLC_INTERFACE=eth0 \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             HF_TOKEN=******                                                                       \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026619;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026620;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m HF_TOKEN=****** \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             PKG_CMD=yum                                                                           \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026627;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026628;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m PKG_CMD=yum \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             PYTHONIOENCODING=UTF-8                                                                \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026635;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026636;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m PYTHONIOENCODING=UTF-\u001b[1m8\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SHLVL=1                                                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026643;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026644;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SHLVL=\u001b[1m1\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             NV_CUDA_LIB_VERSION=13.0.2-1                                                          \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026651;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026652;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m NV_CUDA_LIB_VERSION=\u001b[1m13\u001b[0m\u001b[1m.0\u001b[0m.\u001b[1m2\u001b[0m-\u001b[1m1\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             NVARCH=x86_64                                                                         \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026659;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026660;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m NVARCH=x86_64 \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             PYTHONDONTWRITEBYTECODE=1                                                             \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026667;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026668;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m PYTHONDONTWRITEBYTECODE=\u001b[1m1\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             LD_LIBRARY_PATH=/opt/venv/lib/python3.12/site-packages/nvidia/cudnn                   \n",
+       "                             /lib:/opt/venv/lib/python3.12/site-packages/nvidia/nccl/lib:/opt/am                   \n",
+       "                             azon/ofi-nccl/lib64:/opt/amazon/openmpi/lib:/opt/amazon/openmpi/lib                   \n",
+       "                             64:/opt/amazon/efa/lib:/opt/amazon/efa/lib64:/usr/local/cuda/lib64:                   \n",
+       "                             /usr/local/lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/l                   \n",
+       "                             ocal/cuda/lib64                                                                       \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026675;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026676;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m LD_LIBRARY_PATH=/opt/venv/lib/python3.12/site-packages/nvidia/cudnn \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m /lib:/opt/venv/lib/python3.12/site-packages/nvidia/nccl/lib:/opt/am \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m azon/ofi-nccl/lib64:/opt/amazon/openmpi/lib:/opt/amazon/openmpi/lib \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 64:/opt/amazon/efa/lib:/opt/amazon/efa/lib64:/usr/local/cuda/lib64: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m /usr/local/lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/l \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m ocal/cuda/lib64 \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             TRAINING_JOB_NAME=smolm3-grpo-toolcall-20260626-081812-202606261018                   \n",
+       "                             12                                                                                    \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026683;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026684;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m TRAINING_JOB_NAME=smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m202606261018\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m12\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             LC_ALL=C.UTF-8                                                                        \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026691;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026692;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m LC_ALL=C.UTF-\u001b[1m8\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             TRAINING_JOB_ARN=arn:aws:sagemaker:us-east-1:754289655784:training-                   \n",
+       "                             job/smolm3-grpo-toolcall-20260626-081812-20260626101812                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026699;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026700;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m TRAINING_JOB_ARN=arn:aws:sagemaker:us-east-1:754289655784:training- \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m job/smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             CUDA_HOME=/usr/local/cuda                                                             \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026707;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026708;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m CUDA_HOME=/usr/local/cuda \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             PATH=/opt/venv/bin:/opt/amazon/openmpi/bin:/opt/amazon/efa/bin:/usr                   \n",
+       "                             /local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/loca                   \n",
+       "                             l/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin                                   \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026715;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026716;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m PATH=/opt/venv/bin:/opt/amazon/openmpi/bin:/opt/amazon/efa/bin:/usr \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m /local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/loca \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m l/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             UV_PROJECT_ENVIRONMENT=/opt/venv                                                      \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026723;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026724;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m UV_PROJECT_ENVIRONMENT=/opt/venv \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             DLC_CONTAINER_TYPE=training                                                           \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026731;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026732;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m DLC_CONTAINER_TYPE=training \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             _=/opt/venv/bin/python3                                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026739;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026740;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m _=/opt/venv/bin/python3 \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_MODEL_DIR=/opt/ml/model                                                            \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026747;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026748;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_MODEL_DIR=/opt/ml/model \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_INPUT_DIR=/opt/ml/input                                                            \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026755;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026756;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_IN\u001b[1mPUT\u001b[0m_DIR=/opt/ml/input \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_INPUT_DATA_DIR=/opt/ml/input/data                                                  \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026763;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026764;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_IN\u001b[1mPUT\u001b[0m_DATA_DIR=/opt/ml/input/data \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_INPUT_CONFIG_DIR=/opt/ml/input/config                                              \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026771;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026772;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_IN\u001b[1mPUT\u001b[0m_CONFIG_DIR=/opt/ml/input/config \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_OUTPUT_DIR=/opt/ml/output                                                          \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026779;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026780;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_OUT\u001b[1mPUT\u001b[0m_DIR=/opt/ml/output \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_OUTPUT_FAILURE=/opt/ml/output/failure                                              \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026787;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026788;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_OUT\u001b[1mPUT\u001b[0m_FAILURE=/opt/ml/output/failure \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_OUTPUT_DATA_DIR=/opt/ml/output/data                                                \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026795;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026796;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_OUT\u001b[1mPUT\u001b[0m_DATA_DIR=/opt/ml/output/data \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_LOG_LEVEL=20                                                                       \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026803;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026804;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_LOG_LEVEL=\u001b[1m20\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_MASTER_ADDR=algo-1                                                                 \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026811;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026812;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_MASTER_ADDR=algo-\u001b[1m1\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_MASTER_PORT=7777                                                                   \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026819;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026820;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_MASTER_PORT=\u001b[1m7777\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_SOURCE_DIR=/opt/ml/input/data/code                                                 \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026827;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026828;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_SOURCE_DIR=/opt/ml/input/data/code \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_ENTRY_SCRIPT=launch.sh                                                             \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026835;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026836;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_ENTRY_SCRIPT=launch.sh \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_CHANNEL_CODE=/opt/ml/input/data/code                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026843;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026844;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_CHANNEL_CODE=/opt/ml/input/data/code \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_CHANNEL_SM_DRIVERS=/opt/ml/input/data/sm_drivers                                   \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026851;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026852;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_CHANNEL_SM_DRIVERS=/opt/ml/input/data/sm_drivers \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_CHANNEL_TRAIN=/opt/ml/input/data/train                                             \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026859;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026860;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_CHANNEL_TRAIN=/opt/ml/input/data/train \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_CHANNELS=['code', 'sm_drivers', 'train']                                           \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026867;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026868;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_CHANNELS=\u001b[1m[\u001b[0m'code', 'sm_drivers', 'train'\u001b[1m]\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_HP_BETA=0.1                                                                        \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026875;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026876;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_HP_BETA=\u001b[1m0\u001b[0m\u001b[1m.1\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_HP_DEEPSPEED=ds_zero3.json                                                         \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026883;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026884;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_HP_DEEPSPEED=ds_zero3.json \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_HP_GRADIENT_ACCUMULATION_STEPS=8                                                   \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026891;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026892;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_HP_GRADIENT_ACCUMULATION_STEPS=\u001b[1m8\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_HP_GRADIENT_CHECKPOINTING=True                                                     \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026899;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026900;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_HP_GRADIENT_CHECKPOINTING=\u001b[3mTrue\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_HP_LEARNING_RATE=1e-08                                                             \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026907;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026908;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_HP_LEARNING_RATE=\u001b[1m1e\u001b[0m\u001b[1m-08\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_HP_LOG_COMPLETIONS=False                                                           \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026915;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026916;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_HP_LOG_COMPLETIONS=\u001b[3mFalse\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_HP_MAX_COMPLETION_LENGTH=128                                                       \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026923;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026924;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_HP_MAX_COMPLETION_LENGTH=\u001b[1m128\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_HP_MAX_GRAD_NORM=0.05                                                              \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026931;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026932;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_HP_MAX_GRAD_NORM=\u001b[1m0\u001b[0m\u001b[1m.05\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_HP_MAX_STEPS=50                                                                    \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026939;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026940;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_HP_MAX_STEPS=\u001b[1m50\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_HP_MODEL_ID=HuggingFaceTB/SmolLM3-3B                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026947;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026948;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_HP_MODEL_ID=HuggingFaceTB/SmolLM3-3B \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_HP_NUM_COMPLETIONS_TO_PRINT=8                                                      \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026955;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026956;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_HP_NUM_COMPLETIONS_TO_PRINT=\u001b[1m8\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_HP_NUM_GENERATIONS=8                                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026963;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026964;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_HP_NUM_GENERATIONS=\u001b[1m8\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_HP_PER_DEVICE_TRAIN_BATCH_SIZE=2                                                   \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026971;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026972;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_HP_PER_DEVICE_TRAIN_BATCH_SIZE=\u001b[1m2\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_HP_REMOVE_INVALID_VALUES=False                                                     \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026979;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026980;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_HP_REMOVE_INVALID_VALUES=\u001b[3mFalse\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_HP_RENORMALIZE_LOGITS=True                                                         \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026987;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026988;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_HP_RENORMALIZE_LOGITS=\u001b[3mTrue\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_HP_REPETITION_PENALTY=1.05                                                         \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026995;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026996;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_HP_REPETITION_PENALTY=\u001b[1m1\u001b[0m\u001b[1m.05\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_HP_REWARD_WEIGHTS=1.0,0.5                                                          \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027003;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027004;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_HP_REWARD_WEIGHTS=\u001b[1m1\u001b[0m\u001b[1m.0\u001b[0m,\u001b[1m0.5\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_HP_STOP_ON_COLLAPSE=True                                                           \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027011;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027012;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_HP_STOP_ON_COLLAPSE=\u001b[3mTrue\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_HP_TEMPERATURE=0.7                                                                 \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027019;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027020;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_HP_TEMPERATURE=\u001b[1m0\u001b[0m\u001b[1m.7\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_HP_TOP_K=0                                                                         \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027027;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027028;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_HP_TOP_K=\u001b[1m0\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_HP_TOP_P=1.0                                                                       \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027035;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027036;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_HP_TOP_P=\u001b[1m1\u001b[0m\u001b[1m.0\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_HPS={\"beta\": 0.1, \"deepspeed\": \"ds_zero3.json\",                                    \n",
+       "                             \"gradient_accumulation_steps\": 8, \"gradient_checkpointing\": true,                     \n",
+       "                             \"learning_rate\": 1e-08, \"log_completions\": false,                                     \n",
+       "                             \"max_completion_length\": 128, \"max_grad_norm\": 0.05, \"max_steps\":                     \n",
+       "                             50, \"model_id\": \"HuggingFaceTB/SmolLM3-3B\",                                           \n",
+       "                             \"num_completions_to_print\": 8, \"num_generations\": 8,                                  \n",
+       "                             \"per_device_train_batch_size\": 2, \"remove_invalid_values\": false,                     \n",
+       "                             \"renormalize_logits\": true, \"repetition_penalty\": 1.05,                               \n",
+       "                             \"reward_weights\": \"1.0,0.5\", \"stop_on_collapse\": true,                                \n",
+       "                             \"temperature\": 0.7, \"top_k\": 0, \"top_p\": 1.0}                                         \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027043;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027044;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_HPS=\u001b[1m{\u001b[0m\"beta\": \u001b[1m0.1\u001b[0m, \"deepspeed\": \"ds_zero3.json\", \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"gradient_accumulation_steps\": \u001b[1m8\u001b[0m, \"gradient_checkpointing\": true, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"learning_rate\": \u001b[1m1e-08\u001b[0m, \"log_completions\": false, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"max_completion_length\": \u001b[1m128\u001b[0m, \"max_grad_norm\": \u001b[1m0.05\u001b[0m, \"max_steps\": \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m50\u001b[0m, \"model_id\": \"HuggingFaceTB/SmolLM3-3B\", \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"num_completions_to_print\": \u001b[1m8\u001b[0m, \"num_generations\": \u001b[1m8\u001b[0m, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"per_device_train_batch_size\": \u001b[1m2\u001b[0m, \"remove_invalid_values\": false, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"renormalize_logits\": true, \"repetition_penalty\": \u001b[1m1.05\u001b[0m, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"reward_weights\": \"1.0,0.5\", \"stop_on_collapse\": true, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"temperature\": \u001b[1m0.7\u001b[0m, \"top_k\": \u001b[1m0\u001b[0m, \"top_p\": \u001b[1m1.0\u001b[0m\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_CURRENT_HOST=algo-1                                                                \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027051;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027052;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_CURRENT_HOST=algo-\u001b[1m1\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_CURRENT_INSTANCE_TYPE=ml.p4d.24xlarge                                              \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027059;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027060;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_CURRENT_INSTANCE_TYPE=ml.p4d.24xlarge \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_HOSTS=['algo-1']                                                                   \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027067;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027068;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_HOSTS=\u001b[1m[\u001b[0m'algo-1'\u001b[1m]\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_NETWORK_INTERFACE_NAME=eth0                                                        \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027075;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027076;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_NETWORK_INTERFACE_NAME=eth0 \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_HOST_COUNT=1                                                                       \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027083;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027084;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_HOST_COUNT=\u001b[1m1\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_CURRENT_HOST_RANK=0                                                                \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027091;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027092;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_CURRENT_HOST_RANK=\u001b[1m0\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_NUM_CPUS=96                                                                        \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027099;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027100;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_NUM_CPUS=\u001b[1m96\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_NUM_GPUS=8                                                                         \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027107;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027108;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_NUM_GPUS=\u001b[1m8\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_NUM_NEURONS=0                                                                      \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027115;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027116;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_NUM_NEURONS=\u001b[1m0\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_RESOURCE_CONFIG={\"current_host\": \"algo-1\",                                         \n",
+       "                             \"current_instance_type\": \"ml.p4d.24xlarge\", \"current_group_name\":                     \n",
+       "                             \"homogeneousCluster\", \"hosts\": [\"algo-1\"], \"instance_groups\":                         \n",
+       "                             [{\"instance_group_name\": \"homogeneousCluster\", \"instance_type\":                       \n",
+       "                             \"ml.p4d.24xlarge\", \"hosts\": [\"algo-1\"]}], \"network_interface_name\":                   \n",
+       "                             \"eth0\", \"topology\": null}                                                             \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027123;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027124;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_RESOURCE_CONFIG=\u001b[1m{\u001b[0m\"current_host\": \"algo-1\", \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"current_instance_type\": \"ml.p4d.24xlarge\", \"current_group_name\": \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"homogeneousCluster\", \"hosts\": \u001b[1m[\u001b[0m\"algo-1\"\u001b[1m]\u001b[0m, \"instance_groups\": \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m[\u001b[0m\u001b[1m{\u001b[0m\"instance_group_name\": \"homogeneousCluster\", \"instance_type\": \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"ml.p4d.24xlarge\", \"hosts\": \u001b[1m[\u001b[0m\"algo-1\"\u001b[1m]\u001b[0m\u001b[1m}\u001b[0m\u001b[1m]\u001b[0m, \"network_interface_name\": \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"eth0\", \"topology\": null\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_INPUT_DATA_CONFIG={\"code\": {\"TrainingInputMode\": \"File\",                           \n",
+       "                             \"S3DistributionType\": \"FullyReplicated\", \"RecordWrapperType\":                         \n",
+       "                             \"None\"}, \"sm_drivers\": {\"TrainingInputMode\": \"File\",                                  \n",
+       "                             \"S3DistributionType\": \"FullyReplicated\", \"RecordWrapperType\":                         \n",
+       "                             \"None\"}, \"train\": {\"TrainingInputMode\": \"File\",                                       \n",
+       "                             \"S3DistributionType\": \"FullyReplicated\", \"RecordWrapperType\":                         \n",
+       "                             \"None\"}}                                                                              \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027131;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027132;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_IN\u001b[1mPUT\u001b[0m_DATA_CONFIG=\u001b[1m{\u001b[0m\"code\": \u001b[1m{\u001b[0m\"TrainingInputMode\": \"File\", \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"S3DistributionType\": \"FullyReplicated\", \"RecordWrapperType\": \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"None\"\u001b[1m}\u001b[0m, \"sm_drivers\": \u001b[1m{\u001b[0m\"TrainingInputMode\": \"File\", \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"S3DistributionType\": \"FullyReplicated\", \"RecordWrapperType\": \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"None\"\u001b[1m}\u001b[0m, \"train\": \u001b[1m{\u001b[0m\"TrainingInputMode\": \"File\", \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"S3DistributionType\": \"FullyReplicated\", \"RecordWrapperType\": \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"None\"\u001b[1m}\u001b[0m\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             SM_TRAINING_ENV={\"channel_input_dirs\": {\"code\":                                       \n",
+       "                             \"/opt/ml/input/data/code\", \"sm_drivers\":                                              \n",
+       "                             \"/opt/ml/input/data/sm_drivers\", \"train\":                                             \n",
+       "                             \"/opt/ml/input/data/train\"}, \"current_host\": \"algo-1\",                                \n",
+       "                             \"current_instance_type\": \"ml.p4d.24xlarge\", \"hosts\": [\"algo-1\"],                      \n",
+       "                             \"master_addr\": \"algo-1\", \"master_port\": 7777, \"hyperparameters\":                      \n",
+       "                             {\"beta\": 0.1, \"deepspeed\": \"ds_zero3.json\",                                           \n",
+       "                             \"gradient_accumulation_steps\": 8, \"gradient_checkpointing\": true,                     \n",
+       "                             \"learning_rate\": 1e-08, \"log_completions\": false,                                     \n",
+       "                             \"max_completion_length\": 128, \"max_grad_norm\": 0.05, \"max_steps\":                     \n",
+       "                             50, \"model_id\": \"HuggingFaceTB/SmolLM3-3B\",                                           \n",
+       "                             \"num_completions_to_print\": 8, \"num_generations\": 8,                                  \n",
+       "                             \"per_device_train_batch_size\": 2, \"remove_invalid_values\": false,                     \n",
+       "                             \"renormalize_logits\": true, \"repetition_penalty\": 1.05,                               \n",
+       "                             \"reward_weights\": \"1.0,0.5\", \"stop_on_collapse\": true,                                \n",
+       "                             \"temperature\": 0.7, \"top_k\": 0, \"top_p\": 1.0}, \"input_data_config\":                   \n",
+       "                             {\"code\": {\"TrainingInputMode\": \"File\", \"S3DistributionType\":                          \n",
+       "                             \"FullyReplicated\", \"RecordWrapperType\": \"None\"}, \"sm_drivers\":                        \n",
+       "                             {\"TrainingInputMode\": \"File\", \"S3DistributionType\":                                   \n",
+       "                             \"FullyReplicated\", \"RecordWrapperType\": \"None\"}, \"train\":                             \n",
+       "                             {\"TrainingInputMode\": \"File\", \"S3DistributionType\":                                   \n",
+       "                             \"FullyReplicated\", \"RecordWrapperType\": \"None\"}},                                     \n",
+       "                             \"input_config_dir\": \"/opt/ml/input/config\", \"input_data_dir\":                         \n",
+       "                             \"/opt/ml/input/data\", \"input_dir\": \"/opt/ml/input\", \"job_name\":                       \n",
+       "                             \"smolm3-grpo-toolcall-20260626-081812-20260626101812\", \"log_level\":                   \n",
+       "                             20, \"model_dir\": \"/opt/ml/model\", \"network_interface_name\": \"eth0\",                   \n",
+       "                             \"num_cpus\": 96, \"num_gpus\": 8, \"num_neurons\": 0, \"output_data_dir\":                   \n",
+       "                             \"/opt/ml/output/data\", \"resource_config\": {\"current_host\":                            \n",
+       "                             \"algo-1\", \"current_instance_type\": \"ml.p4d.24xlarge\",                                 \n",
+       "                             \"current_group_name\": \"homogeneousCluster\", \"hosts\": [\"algo-1\"],                      \n",
+       "                             \"instance_groups\": [{\"instance_group_name\": \"homogeneousCluster\",                     \n",
+       "                             \"instance_type\": \"ml.p4d.24xlarge\", \"hosts\": [\"algo-1\"]}],                            \n",
+       "                             \"network_interface_name\": \"eth0\", \"topology\": null}}                                  \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027139;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027140;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m SM_TRAINING_ENV=\u001b[1m{\u001b[0m\"channel_input_dirs\": \u001b[1m{\u001b[0m\"code\": \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"/opt/ml/input/data/code\", \"sm_drivers\": \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"/opt/ml/input/data/sm_drivers\", \"train\": \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"/opt/ml/input/data/train\"\u001b[1m}\u001b[0m, \"current_host\": \"algo-1\", \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"current_instance_type\": \"ml.p4d.24xlarge\", \"hosts\": \u001b[1m[\u001b[0m\"algo-1\"\u001b[1m]\u001b[0m, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"master_addr\": \"algo-1\", \"master_port\": \u001b[1m7777\u001b[0m, \"hyperparameters\": \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m{\u001b[0m\"beta\": \u001b[1m0.1\u001b[0m, \"deepspeed\": \"ds_zero3.json\", \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"gradient_accumulation_steps\": \u001b[1m8\u001b[0m, \"gradient_checkpointing\": true, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"learning_rate\": \u001b[1m1e-08\u001b[0m, \"log_completions\": false, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"max_completion_length\": \u001b[1m128\u001b[0m, \"max_grad_norm\": \u001b[1m0.05\u001b[0m, \"max_steps\": \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m50\u001b[0m, \"model_id\": \"HuggingFaceTB/SmolLM3-3B\", \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"num_completions_to_print\": \u001b[1m8\u001b[0m, \"num_generations\": \u001b[1m8\u001b[0m, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"per_device_train_batch_size\": \u001b[1m2\u001b[0m, \"remove_invalid_values\": false, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"renormalize_logits\": true, \"repetition_penalty\": \u001b[1m1.05\u001b[0m, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"reward_weights\": \"1.0,0.5\", \"stop_on_collapse\": true, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"temperature\": \u001b[1m0.7\u001b[0m, \"top_k\": \u001b[1m0\u001b[0m, \"top_p\": \u001b[1m1.0\u001b[0m\u001b[1m}\u001b[0m, \"input_data_config\": \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m{\u001b[0m\"code\": \u001b[1m{\u001b[0m\"TrainingInputMode\": \"File\", \"S3DistributionType\": \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"FullyReplicated\", \"RecordWrapperType\": \"None\"\u001b[1m}\u001b[0m, \"sm_drivers\": \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m{\u001b[0m\"TrainingInputMode\": \"File\", \"S3DistributionType\": \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"FullyReplicated\", \"RecordWrapperType\": \"None\"\u001b[1m}\u001b[0m, \"train\": \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m{\u001b[0m\"TrainingInputMode\": \"File\", \"S3DistributionType\": \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"FullyReplicated\", \"RecordWrapperType\": \"None\"\u001b[1m}\u001b[0m\u001b[1m}\u001b[0m, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"input_config_dir\": \"/opt/ml/input/config\", \"input_data_dir\": \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"/opt/ml/input/data\", \"input_dir\": \"/opt/ml/input\", \"job_name\": \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"smolm3-grpo-toolcall-20260626-081812-20260626101812\", \"log_level\": \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m20\u001b[0m, \"model_dir\": \"/opt/ml/model\", \"network_interface_name\": \"eth0\", \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"num_cpus\": \u001b[1m96\u001b[0m, \"num_gpus\": \u001b[1m8\u001b[0m, \"num_neurons\": \u001b[1m0\u001b[0m, \"output_data_dir\": \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"/opt/ml/output/data\", \"resource_config\": \u001b[1m{\u001b[0m\"current_host\": \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"algo-1\", \"current_instance_type\": \"ml.p4d.24xlarge\", \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"current_group_name\": \"homogeneousCluster\", \"hosts\": \u001b[1m[\u001b[0m\"algo-1\"\u001b[1m]\u001b[0m, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"instance_groups\": \u001b[1m[\u001b[0m\u001b[1m{\u001b[0m\"instance_group_name\": \"homogeneousCluster\", \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"instance_type\": \"ml.p4d.24xlarge\", \"hosts\": \u001b[1m[\u001b[0m\"algo-1\"\u001b[1m]\u001b[0m\u001b[1m}\u001b[0m\u001b[1m]\u001b[0m, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \"network_interface_name\": \"eth0\", \"topology\": null\u001b[1m}\u001b[0m\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             ++ set +x                                                                             \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027147;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027148;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m ++ set +x \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             ++ cd /opt/ml/input/data/code                                                         \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027155;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027156;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m ++ cd /opt/ml/input/data/code \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             Running Basic Script driver                                                           \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027163;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027164;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Running Basic Script driver \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             ++ echo 'Running Basic Script driver'                                                 \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027171;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027172;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m ++ echo 'Running Basic Script driver' \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             ++ /opt/venv/bin/python3                                                              \n",
+       "                             /opt/ml/input/data/sm_drivers/distributed_drivers/basic_script_driv                   \n",
+       "                             er.py                                                                                 \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027179;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027180;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m ++ /opt/venv/bin/python3 \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m /opt/ml/input/data/sm_drivers/distributed_drivers/basic_script_driv \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m er.py \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             Executing command: /bin/sh -c chmod +x launch.sh && ./launch.sh                       \n",
+       "                             --beta 0.1 --deepspeed ds_zero3.json --gradient_accumulation_steps                    \n",
+       "                             8 --gradient_checkpointing true --learning_rate 1e-08                                 \n",
+       "                             --log_completions false --max_completion_length 128 --max_grad_norm                   \n",
+       "                             0.05 --max_steps 50 --model_id HuggingFaceTB/SmolLM3-3B                               \n",
+       "                             --num_completions_to_print 8 --num_generations 8                                      \n",
+       "                             --per_device_train_batch_size 2 --remove_invalid_values false                         \n",
+       "                             --renormalize_logits true --repetition_penalty 1.05                                   \n",
+       "                             --reward_weights 1.0,0.5 --stop_on_collapse true --temperature 0.7                    \n",
+       "                             --top_k 0 --top_p 1.0                                                                 \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027187;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027188;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Executing command: /bin/sh -c chmod +x launch.sh && ./launch.sh \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m --beta \u001b[1m0.1\u001b[0m --deepspeed ds_zero3.json --gradient_accumulation_steps \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m8\u001b[0m --gradient_checkpointing true --learning_rate \u001b[1m1e-08\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m --log_completions false --max_completion_length \u001b[1m128\u001b[0m --max_grad_norm \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m0.05\u001b[0m --max_steps \u001b[1m50\u001b[0m --model_id HuggingFaceTB/SmolLM3-3B \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m --num_completions_to_print \u001b[1m8\u001b[0m --num_generations \u001b[1m8\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m --per_device_train_batch_size \u001b[1m2\u001b[0m --remove_invalid_values false \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m --renormalize_logits true --repetition_penalty \u001b[1m1.05\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m --reward_weights \u001b[1m1.0\u001b[0m,\u001b[1m0.5\u001b[0m --stop_on_collapse true --temperature \u001b[1m0.7\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m --top_k \u001b[1m0\u001b[0m --top_p \u001b[1m1.0\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             W0626 08:29:15.449000 26 torch/distributed/run.py:851]                                \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027195;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027196;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m W0626 08:29:15.\u001b[1m449000\u001b[0m \u001b[1m26\u001b[0m torch/distributed/run.py:\u001b[1m851\u001b[0m\u001b[1m]\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             W0626 08:29:15.449000 26 torch/distributed/run.py:851]                                \n",
+       "                             *****************************************                                             \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027203;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027204;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m W0626 08:29:15.\u001b[1m449000\u001b[0m \u001b[1m26\u001b[0m torch/distributed/run.py:\u001b[1m851\u001b[0m\u001b[1m]\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m ***************************************** \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             W0626 08:29:15.449000 26 torch/distributed/run.py:851] Setting                        \n",
+       "                             OMP_NUM_THREADS environment variable for each process to be 1 in                      \n",
+       "                             default, to avoid your system being overloaded, please further tune                   \n",
+       "                             the variable for optimal performance in your application as needed.                   \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027211;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027212;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m W0626 08:29:15.\u001b[1m449000\u001b[0m \u001b[1m26\u001b[0m torch/distributed/run.py:\u001b[1m851\u001b[0m\u001b[1m]\u001b[0m Setting \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m OMP_NUM_THREADS environment variable for each process to be \u001b[1m1\u001b[0m in \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m default, to avoid your system being overloaded, please further tune \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m the variable for optimal performance in your application as needed. \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             W0626 08:29:15.449000 26 torch/distributed/run.py:851]                                \n",
+       "                             *****************************************                                             \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027219;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027220;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m W0626 08:29:15.\u001b[1m449000\u001b[0m \u001b[1m26\u001b[0m torch/distributed/run.py:\u001b[1m851\u001b[0m\u001b[1m]\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m ***************************************** \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:29:57] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             df: /root/.triton/autotune: No such file or directory                                 \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:29:57]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027227;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027228;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m df: /root/.triton/autotune: No such file or directory \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             NCCL version 2.28.9+cuda13.0                                                          \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027235;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027236;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m NCCL version \u001b[1m2.28\u001b[0m.\u001b[1m9\u001b[0m+cuda13.\u001b[1m0\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:30:07] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             Fetching 2 files:   0%|          | 0/2 [00:00<?, ?it/s]#015Fetching                   \n",
+       "                             2 files:   0%|          | 0/2 [00:00<?, ?it/s]#015Fetching 2 files:                   \n",
+       "                             0%|          | 0/2 [00:00<?, ?it/s]#015Fetching 2 files:   0%|                        \n",
+       "                             | 0/2 [00:00<?, ?it/s]#015Fetching 2 files:   0%|          | 0/2                      \n",
+       "                             [00:00<?, ?it/s]#015Fetching 2 files:   0%|          | 0/2                            \n",
+       "                             [00:00<?, ?it/s]#015Fetching 2 files:   0%|          | 0/2                            \n",
+       "                             [00:00<?, ?it/s]#015Fetching 2 files:   0%|          | 0/2                            \n",
+       "                             [00:00<?, ?it/s]#015Fetching 2 files:  50%|█████     | 1/2                            \n",
+       "                             [00:08<00:08,  8.77s/it]#015Fetching 2 files: 100%|██████████| 2/2                    \n",
+       "                             [00:08<00:00,  4.39s/it]                                                              \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:30:07]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027243;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027244;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Fetching \u001b[1m2\u001b[0m files: \u001b[1m0\u001b[0m%| | \u001b[1m0\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:00 INFO smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n", + " 09: \n", + " Fetching 2 files: 50%|█████ | 1/2 [00:08<00:08, \n", + " 8.76s/it]#015Fetching 2 files: 100%|██████████| 2/2 [00:08<00:00, \n", + " 4.38s/it] \n", + "\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027251;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027252;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Fetching \u001b[1m2\u001b[0m files: \u001b[1m50\u001b[0m%|█████ | \u001b[1m1\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:08<00:08, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m8.\u001b[0m76s/it\u001b[1m]\u001b[0m#015Fetching \u001b[1m2\u001b[0m files: \u001b[1m100\u001b[0m%|██████████| \u001b[1m2\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:08<00:00, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m4.\u001b[0m38s/it\u001b[1m]\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             Fetching 2 files:  50%|█████     | 1/2 [00:08<00:08,                                  \n",
+       "                             8.76s/it]#015Fetching 2 files: 100%|██████████| 2/2 [00:08<00:00,                     \n",
+       "                             4.38s/it]                                                                             \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027259;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027260;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Fetching \u001b[1m2\u001b[0m files: \u001b[1m50\u001b[0m%|█████ | \u001b[1m1\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:08<00:08, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m8.\u001b[0m76s/it\u001b[1m]\u001b[0m#015Fetching \u001b[1m2\u001b[0m files: \u001b[1m100\u001b[0m%|██████████| \u001b[1m2\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:08<00:00, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m4.\u001b[0m38s/it\u001b[1m]\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             Fetching 2 files:  50%|█████     | 1/2 [00:08<00:08,                                  \n",
+       "                             8.71s/it]#015Fetching 2 files: 100%|██████████| 2/2 [00:08<00:00,                     \n",
+       "                             4.36s/it]                                                                             \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027267;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027268;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Fetching \u001b[1m2\u001b[0m files: \u001b[1m50\u001b[0m%|█████ | \u001b[1m1\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:08<00:08, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m8.\u001b[0m71s/it\u001b[1m]\u001b[0m#015Fetching \u001b[1m2\u001b[0m files: \u001b[1m100\u001b[0m%|██████████| \u001b[1m2\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:08<00:00, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m4.\u001b[0m36s/it\u001b[1m]\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             Fetching 2 files:  50%|█████     | 1/2 [00:08<00:08,                                  \n",
+       "                             8.77s/it]#015Fetching 2 files: 100%|██████████| 2/2 [00:08<00:00,                     \n",
+       "                             4.38s/it]                                                                             \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027275;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027276;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Fetching \u001b[1m2\u001b[0m files: \u001b[1m50\u001b[0m%|█████ | \u001b[1m1\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:08<00:08, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m8.\u001b[0m77s/it\u001b[1m]\u001b[0m#015Fetching \u001b[1m2\u001b[0m files: \u001b[1m100\u001b[0m%|██████████| \u001b[1m2\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:08<00:00, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m4.\u001b[0m38s/it\u001b[1m]\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             Fetching 2 files:  50%|█████     | 1/2 [00:08<00:08,                                  \n",
+       "                             8.77s/it]#015Fetching 2 files: 100%|██████████| 2/2 [00:08<00:00,                     \n",
+       "                             4.38s/it]                                                                             \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027283;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027284;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Fetching \u001b[1m2\u001b[0m files: \u001b[1m50\u001b[0m%|█████ | \u001b[1m1\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:08<00:08, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m8.\u001b[0m77s/it\u001b[1m]\u001b[0m#015Fetching \u001b[1m2\u001b[0m files: \u001b[1m100\u001b[0m%|██████████| \u001b[1m2\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:08<00:00, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m4.\u001b[0m38s/it\u001b[1m]\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             Fetching 2 files:  50%|█████     | 1/2 [00:08<00:08,                                  \n",
+       "                             8.76s/it]#015Fetching 2 files: 100%|██████████| 2/2 [00:08<00:00,                     \n",
+       "                             4.38s/it]                                                                             \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027291;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027292;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Fetching \u001b[1m2\u001b[0m files: \u001b[1m50\u001b[0m%|█████ | \u001b[1m1\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:08<00:08, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m8.\u001b[0m76s/it\u001b[1m]\u001b[0m#015Fetching \u001b[1m2\u001b[0m files: \u001b[1m100\u001b[0m%|██████████| \u001b[1m2\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:08<00:00, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m4.\u001b[0m38s/it\u001b[1m]\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             Fetching 2 files:  50%|█████     | 1/2 [00:08<00:08,                                  \n",
+       "                             8.82s/it]#015Fetching 2 files: 100%|██████████| 2/2 [00:08<00:00,                     \n",
+       "                             4.41s/it]                                                                             \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027299;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027300;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Fetching \u001b[1m2\u001b[0m files: \u001b[1m50\u001b[0m%|█████ | \u001b[1m1\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:08<00:08, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m8.\u001b[0m82s/it\u001b[1m]\u001b[0m#015Fetching \u001b[1m2\u001b[0m files: \u001b[1m100\u001b[0m%|██████████| \u001b[1m2\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:08<00:00, \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m4.\u001b[0m41s/it\u001b[1m]\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             Fetching 2 files:   0%|          | 0/2 [00:00<?, ?it/s]#015Fetching                   \n",
+       "                             2 files: 100%|██████████| 2/2 [00:00<00:00, 44150.57it/s]                             \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027307;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027308;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Fetching \u001b[1m2\u001b[0m files: \u001b[1m0\u001b[0m%| | \u001b[1m0\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:00 INFO smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n", + " 09: \n", + " Fetching 2 files: 0%| | 0/2 [00:00<?, ?it/s]#015Fetching \n", + " 2 files: 100%|██████████| 2/2 [00:00<00:00, 41734.37it/s] \n", + "\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027315;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027316;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Fetching \u001b[1m2\u001b[0m files: \u001b[1m0\u001b[0m%| | \u001b[1m0\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:00 INFO smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n", + " 09: \n", + " Fetching 2 files: 0%| | 0/2 [00:00<?, ?it/s]#015Fetching \n", + " 2 files: 100%|██████████| 2/2 [00:00<00:00, 43690.67it/s] \n", + "\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027323;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027324;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Fetching \u001b[1m2\u001b[0m files: \u001b[1m0\u001b[0m%| | \u001b[1m0\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:00 INFO smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n", + " 09: \n", + " Fetching 2 files: 0%| | 0/2 [00:00<?, ?it/s]#015Fetching \n", + " 2 files: 100%|██████████| 2/2 [00:00<00:00, 40329.85it/s] \n", + "\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027331;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027332;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Fetching \u001b[1m2\u001b[0m files: \u001b[1m0\u001b[0m%| | \u001b[1m0\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:00 INFO smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n", + " 09: \n", + " Fetching 2 files: 0%| | 0/2 [00:00<?, ?it/s]#015Fetching \n", + " 2 files: 100%|██████████| 2/2 [00:00<00:00, 37117.73it/s] \n", + "\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027339;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027340;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Fetching \u001b[1m2\u001b[0m files: \u001b[1m0\u001b[0m%| | \u001b[1m0\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:00 INFO smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n", + " 09: \n", + " Fetching 2 files: 0%| | 0/2 [00:00<?, ?it/s]#015Fetching \n", + " 2 files: 100%|██████████| 2/2 [00:00<00:00, 43018.50it/s] \n", + "\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027347;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027348;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Fetching \u001b[1m2\u001b[0m files: \u001b[1m0\u001b[0m%| | \u001b[1m0\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:00 INFO smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n", + " 09: \n", + " Fetching 2 files: 0%| | 0/2 [00:00<?, ?it/s]#015Fetching \n", + " 2 files: 100%|██████████| 2/2 [00:00<00:00, 41120.63it/s] \n", + "\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027355;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027356;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Fetching \u001b[1m2\u001b[0m files: \u001b[1m0\u001b[0m%| | \u001b[1m0\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:00 INFO smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n", + " 09: \n", + " Fetching 2 files: 0%| | 0/2 [00:00<?, ?it/s]#015Fetching \n", + " 2 files: 100%|██████████| 2/2 [00:00<00:00, 39945.75it/s] \n", + "\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027363;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027364;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Fetching \u001b[1m2\u001b[0m files: \u001b[1m0\u001b[0m%| | \u001b[1m0\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:00[06/26/26 10:30:14] INFO smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n", + " 09: \n", + " The tokenizer has new PAD/BOS/EOS tokens that differ from the \n", + " model config and generation config. The model config and generation \n", + " config were aligned accordingly, being updated with the tokenizer's \n", + " values. Updated tokens: {'bos_token_id': None, 'pad_token_id': \n", + " 128012}. \n", + "\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:30:14]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027371;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027372;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m The tokenizer has new PAD/BOS/EOS tokens that differ from the \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m model config and generation config. The model config and generation \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m config were aligned accordingly, being updated with the tokenizer's \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m values. Updated tokens: \u001b[1m{\u001b[0m'bos_token_id': \u001b[3mNone\u001b[0m, 'pad_token_id': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m128012\u001b[0m\u001b[1m}\u001b[0m. \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                              The tokenizer has new PAD/BOS/EOS tokens that differ from the                        \n",
+       "                             model config and generation config. The model config and generation                   \n",
+       "                             config were aligned accordingly, being updated with the tokenizer's                   \n",
+       "                             values. Updated tokens: {'bos_token_id': None, 'pad_token_id':                        \n",
+       "                             128012}.                                                                              \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027379;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027380;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m The tokenizer has new PAD/BOS/EOS tokens that differ from the \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m model config and generation config. The model config and generation \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m config were aligned accordingly, being updated with the tokenizer's \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m values. Updated tokens: \u001b[1m{\u001b[0m'bos_token_id': \u001b[3mNone\u001b[0m, 'pad_token_id': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m128012\u001b[0m\u001b[1m}\u001b[0m. \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                              The tokenizer has new PAD/BOS/EOS tokens that differ from the                        \n",
+       "                             model config and generation config. The model config and generation                   \n",
+       "                             config were aligned accordingly, being updated with the tokenizer's                   \n",
+       "                             values. Updated tokens: {'bos_token_id': None, 'pad_token_id':                        \n",
+       "                             128012}.                                                                              \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027387;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027388;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m The tokenizer has new PAD/BOS/EOS tokens that differ from the \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m model config and generation config. The model config and generation \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m config were aligned accordingly, being updated with the tokenizer's \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m values. Updated tokens: \u001b[1m{\u001b[0m'bos_token_id': \u001b[3mNone\u001b[0m, 'pad_token_id': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m128012\u001b[0m\u001b[1m}\u001b[0m. \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                              The tokenizer has new PAD/BOS/EOS tokens that differ from the                        \n",
+       "                             model config and generation config. The model config and generation                   \n",
+       "                             config were aligned accordingly, being updated with the tokenizer's                   \n",
+       "                             values. Updated tokens: {'bos_token_id': None, 'pad_token_id':                        \n",
+       "                             128012}.                                                                              \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027395;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027396;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m The tokenizer has new PAD/BOS/EOS tokens that differ from the \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m model config and generation config. The model config and generation \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m config were aligned accordingly, being updated with the tokenizer's \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m values. Updated tokens: \u001b[1m{\u001b[0m'bos_token_id': \u001b[3mNone\u001b[0m, 'pad_token_id': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m128012\u001b[0m\u001b[1m}\u001b[0m. \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                              The tokenizer has new PAD/BOS/EOS tokens that differ from the                        \n",
+       "                             model config and generation config. The model config and generation                   \n",
+       "                             config were aligned accordingly, being updated with the tokenizer's                   \n",
+       "                             values. Updated tokens: {'bos_token_id': None, 'pad_token_id':                        \n",
+       "                             128012}.                                                                              \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027403;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027404;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m The tokenizer has new PAD/BOS/EOS tokens that differ from the \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m model config and generation config. The model config and generation \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m config were aligned accordingly, being updated with the tokenizer's \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m values. Updated tokens: \u001b[1m{\u001b[0m'bos_token_id': \u001b[3mNone\u001b[0m, 'pad_token_id': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m128012\u001b[0m\u001b[1m}\u001b[0m. \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                              The tokenizer has new PAD/BOS/EOS tokens that differ from the                        \n",
+       "                             model config and generation config. The model config and generation                   \n",
+       "                             config were aligned accordingly, being updated with the tokenizer's                   \n",
+       "                             values. Updated tokens: {'bos_token_id': None, 'pad_token_id':                        \n",
+       "                             128012}.                                                                              \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027411;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027412;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m The tokenizer has new PAD/BOS/EOS tokens that differ from the \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m model config and generation config. The model config and generation \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m config were aligned accordingly, being updated with the tokenizer's \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m values. Updated tokens: \u001b[1m{\u001b[0m'bos_token_id': \u001b[3mNone\u001b[0m, 'pad_token_id': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m128012\u001b[0m\u001b[1m}\u001b[0m. \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                              The tokenizer has new PAD/BOS/EOS tokens that differ from the                        \n",
+       "                             model config and generation config. The model config and generation                   \n",
+       "                             config were aligned accordingly, being updated with the tokenizer's                   \n",
+       "                             values. Updated tokens: {'bos_token_id': None, 'pad_token_id':                        \n",
+       "                             128012}.                                                                              \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027419;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027420;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m The tokenizer has new PAD/BOS/EOS tokens that differ from the \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m model config and generation config. The model config and generation \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m config were aligned accordingly, being updated with the tokenizer's \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m values. Updated tokens: \u001b[1m{\u001b[0m'bos_token_id': \u001b[3mNone\u001b[0m, 'pad_token_id': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m128012\u001b[0m\u001b[1m}\u001b[0m. \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                              The tokenizer has new PAD/BOS/EOS tokens that differ from the                        \n",
+       "                             model config and generation config. The model config and generation                   \n",
+       "                             config were aligned accordingly, being updated with the tokenizer's                   \n",
+       "                             values. Updated tokens: {'bos_token_id': None, 'pad_token_id':                        \n",
+       "                             128012}.                                                                              \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027427;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027428;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m The tokenizer has new PAD/BOS/EOS tokens that differ from the \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m model config and generation config. The model config and generation \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m config were aligned accordingly, being updated with the tokenizer's \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m values. Updated tokens: \u001b[1m{\u001b[0m'bos_token_id': \u001b[3mNone\u001b[0m, 'pad_token_id': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m128012\u001b[0m\u001b[1m}\u001b[0m. \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             [RANK 0] Gradient accumulation steps mismatch:                                        \n",
+       "                             GradientAccumulationPlugin has 1, DeepSpeed config has 8. Using                       \n",
+       "                             DeepSpeed's value.                                                                    \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027435;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027436;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m[\u001b[0mRANK \u001b[1m0\u001b[0m\u001b[1m]\u001b[0m Gradient accumulation steps mismatch: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m GradientAccumulationPlugin has \u001b[1m1\u001b[0m, DeepSpeed config has \u001b[1m8\u001b[0m. Using \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m DeepSpeed's value. \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:30:25] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             0%|          | 0/50 [00:00<?, ?it/s] Ignoring                                         \n",
+       "                             clean_up_tokenization_spaces=True for BPE tokenizer                                   \n",
+       "                             TokenizersBackend. The clean_up_tokenization post-processing step                     \n",
+       "                             is designed for WordPiece tokenizers and is destructive for BPE (it                   \n",
+       "                             strips spaces before punctuation). Set                                                \n",
+       "                             clean_up_tokenization_spaces=False to suppress this warning, or set                   \n",
+       "                             clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou                   \n",
+       "                             tput=True to force cleanup anyway.                                                    \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:30:25]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027443;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027444;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m0\u001b[0m%| | \u001b[1m0\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m00:00 INFO smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n", + " 09: \n", + " Ignoring clean_up_tokenization_spaces=True for BPE tokenizer \n", + " TokenizersBackend. The clean_up_tokenization post-processing step \n", + " is designed for WordPiece tokenizers and is destructive for BPE (it \n", + " strips spaces before punctuation). Set \n", + " clean_up_tokenization_spaces=False to suppress this warning, or set \n", + " clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou \n", + " tput=True to force cleanup anyway. \n", + "\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027451;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027452;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Ignoring clean_up_tokenization_spaces=\u001b[3mTrue\u001b[0m for BPE tokenizer \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m TokenizersBackend. The clean_up_tokenization post-processing step \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m is designed for WordPiece tokenizers and is destructive for BPE \u001b[1m(\u001b[0mit \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m strips spaces before punctuation\u001b[1m)\u001b[0m. Set \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m clean_up_tokenization_spaces=\u001b[3mFalse\u001b[0m to suppress this warning, or set \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m tput=\u001b[3mTrue\u001b[0m to force cleanup anyway. \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                              Ignoring clean_up_tokenization_spaces=True for BPE tokenizer                         \n",
+       "                             TokenizersBackend. The clean_up_tokenization post-processing step                     \n",
+       "                             is designed for WordPiece tokenizers and is destructive for BPE (it                   \n",
+       "                             strips spaces before punctuation). Set                                                \n",
+       "                             clean_up_tokenization_spaces=False to suppress this warning, or set                   \n",
+       "                             clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou                   \n",
+       "                             tput=True to force cleanup anyway.                                                    \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027459;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027460;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Ignoring clean_up_tokenization_spaces=\u001b[3mTrue\u001b[0m for BPE tokenizer \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m TokenizersBackend. The clean_up_tokenization post-processing step \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m is designed for WordPiece tokenizers and is destructive for BPE \u001b[1m(\u001b[0mit \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m strips spaces before punctuation\u001b[1m)\u001b[0m. Set \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m clean_up_tokenization_spaces=\u001b[3mFalse\u001b[0m to suppress this warning, or set \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m tput=\u001b[3mTrue\u001b[0m to force cleanup anyway. \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                              Ignoring clean_up_tokenization_spaces=True for BPE tokenizer                         \n",
+       "                             TokenizersBackend. The clean_up_tokenization post-processing step                     \n",
+       "                             is designed for WordPiece tokenizers and is destructive for BPE (it                   \n",
+       "                             strips spaces before punctuation). Set                                                \n",
+       "                             clean_up_tokenization_spaces=False to suppress this warning, or set                   \n",
+       "                             clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou                   \n",
+       "                             tput=True to force cleanup anyway.                                                    \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027467;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027468;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Ignoring clean_up_tokenization_spaces=\u001b[3mTrue\u001b[0m for BPE tokenizer \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m TokenizersBackend. The clean_up_tokenization post-processing step \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m is designed for WordPiece tokenizers and is destructive for BPE \u001b[1m(\u001b[0mit \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m strips spaces before punctuation\u001b[1m)\u001b[0m. Set \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m clean_up_tokenization_spaces=\u001b[3mFalse\u001b[0m to suppress this warning, or set \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m tput=\u001b[3mTrue\u001b[0m to force cleanup anyway. \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                              Ignoring clean_up_tokenization_spaces=True for BPE tokenizer                         \n",
+       "                             TokenizersBackend. The clean_up_tokenization post-processing step                     \n",
+       "                             is designed for WordPiece tokenizers and is destructive for BPE (it                   \n",
+       "                             strips spaces before punctuation). Set                                                \n",
+       "                             clean_up_tokenization_spaces=False to suppress this warning, or set                   \n",
+       "                             clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou                   \n",
+       "                             tput=True to force cleanup anyway.                                                    \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027475;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027476;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Ignoring clean_up_tokenization_spaces=\u001b[3mTrue\u001b[0m for BPE tokenizer \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m TokenizersBackend. The clean_up_tokenization post-processing step \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m is designed for WordPiece tokenizers and is destructive for BPE \u001b[1m(\u001b[0mit \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m strips spaces before punctuation\u001b[1m)\u001b[0m. Set \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m clean_up_tokenization_spaces=\u001b[3mFalse\u001b[0m to suppress this warning, or set \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m tput=\u001b[3mTrue\u001b[0m to force cleanup anyway. \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                              Ignoring clean_up_tokenization_spaces=True for BPE tokenizer                         \n",
+       "                             TokenizersBackend. The clean_up_tokenization post-processing step                     \n",
+       "                             is designed for WordPiece tokenizers and is destructive for BPE (it                   \n",
+       "                             strips spaces before punctuation). Set                                                \n",
+       "                             clean_up_tokenization_spaces=False to suppress this warning, or set                   \n",
+       "                             clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou                   \n",
+       "                             tput=True to force cleanup anyway.                                                    \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027483;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027484;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Ignoring clean_up_tokenization_spaces=\u001b[3mTrue\u001b[0m for BPE tokenizer \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m TokenizersBackend. The clean_up_tokenization post-processing step \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m is designed for WordPiece tokenizers and is destructive for BPE \u001b[1m(\u001b[0mit \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m strips spaces before punctuation\u001b[1m)\u001b[0m. Set \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m clean_up_tokenization_spaces=\u001b[3mFalse\u001b[0m to suppress this warning, or set \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m tput=\u001b[3mTrue\u001b[0m to force cleanup anyway. \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                              Ignoring clean_up_tokenization_spaces=True for BPE tokenizer                         \n",
+       "                             TokenizersBackend. The clean_up_tokenization post-processing step                     \n",
+       "                             is designed for WordPiece tokenizers and is destructive for BPE (it                   \n",
+       "                             strips spaces before punctuation). Set                                                \n",
+       "                             clean_up_tokenization_spaces=False to suppress this warning, or set                   \n",
+       "                             clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou                   \n",
+       "                             tput=True to force cleanup anyway.                                                    \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027491;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027492;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Ignoring clean_up_tokenization_spaces=\u001b[3mTrue\u001b[0m for BPE tokenizer \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m TokenizersBackend. The clean_up_tokenization post-processing step \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m is designed for WordPiece tokenizers and is destructive for BPE \u001b[1m(\u001b[0mit \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m strips spaces before punctuation\u001b[1m)\u001b[0m. Set \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m clean_up_tokenization_spaces=\u001b[3mFalse\u001b[0m to suppress this warning, or set \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m tput=\u001b[3mTrue\u001b[0m to force cleanup anyway. \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                              Ignoring clean_up_tokenization_spaces=True for BPE tokenizer                         \n",
+       "                             TokenizersBackend. The clean_up_tokenization post-processing step                     \n",
+       "                             is designed for WordPiece tokenizers and is destructive for BPE (it                   \n",
+       "                             strips spaces before punctuation). Set                                                \n",
+       "                             clean_up_tokenization_spaces=False to suppress this warning, or set                   \n",
+       "                             clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou                   \n",
+       "                             tput=True to force cleanup anyway.                                                    \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027499;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027500;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Ignoring clean_up_tokenization_spaces=\u001b[3mTrue\u001b[0m for BPE tokenizer \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m TokenizersBackend. The clean_up_tokenization post-processing step \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m is designed for WordPiece tokenizers and is destructive for BPE \u001b[1m(\u001b[0mit \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m strips spaces before punctuation\u001b[1m)\u001b[0m. Set \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m clean_up_tokenization_spaces=\u001b[3mFalse\u001b[0m to suppress this warning, or set \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m tput=\u001b[3mTrue\u001b[0m to force cleanup anyway. \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:30:36] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             2%|▏         | 1/50 [00:19<16:16, 19.92s/it]#015                                      \n",
+       "                             #015{'loss': '0.03134', 'grad_norm': '2.996', 'learning_rate':                        \n",
+       "                             '1e-08', 'num_tokens': '6.144e+04', 'completions/mean_length':                        \n",
+       "                             '30.66', 'completions/min_length': '17', 'completions/max_length':                    \n",
+       "                             '128', 'completions/clipped_ratio': '0.01562',                                        \n",
+       "                             'completions/mean_terminated_length': '29.12',                                        \n",
+       "                             'completions/min_terminated_length': '17',                                            \n",
+       "                             'completions/max_terminated_length': '104',                                           \n",
+       "                             'rewards/tool_call_reward/mean': '0.8281',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.3788',                                             \n",
+       "                             'rewards/format_reward/mean': '0.937', 'rewards/format_reward/std':                   \n",
+       "                             '0.2449', 'reward': '1.297', 'reward_std': '0.4594',                                  \n",
+       "                             'frac_reward_zero_std': '0', 'kl': '4.555e-06', 'entropy':                            \n",
+       "                             '0.02574', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',                     \n",
+       "                             'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',                              \n",
+       "                             'clip_ratio/region_mean': '0', 'step_time': '19.91', 'epoch':                         \n",
+       "                             '0.008'}                                                                              \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:30:36]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027507;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027508;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m2\u001b[0m%|▏ | \u001b[1m1\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m00:19<16:16, \u001b[1m19.\u001b[0m92s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.03134', 'grad_norm': '2.996', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '1e-08', 'num_tokens': '6.144e+04', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '30.66', 'completions/min_length': '17', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '128', 'completions/clipped_ratio': '0.01562', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '29.12', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '17', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '104', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.8281', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.3788', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.937', 'rewards/format_reward/std': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.2449', 'reward': '1.297', 'reward_std': '0.4594', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'frac_reward_zero_std': '0', 'kl': '4.555e-06', 'entropy': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.02574', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/region_mean': '0', 'step_time': '19.91', 'epoch': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.008'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:30:52] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             2%|▏         | 1/50 [00:19<16:16, 19.92s/it]#015  4%|▍         |                      \n",
+       "                             2/50 [00:34<13:36, 17.00s/it]#015                                                     \n",
+       "                             #015{'loss': '0.002899', 'grad_norm': '1.937', 'learning_rate':                       \n",
+       "                             '9.8e-09', 'num_tokens': '1.326e+05', 'completions/mean_length':                      \n",
+       "                             '36.66', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '82', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '36.66',                                        \n",
+       "                             'completions/min_terminated_length': '20',                                            \n",
+       "                             'completions/max_terminated_length': '82',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.6484',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4793',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9371',                                               \n",
+       "                             'rewards/format_reward/std': '0.2445', 'reward': '1.117',                             \n",
+       "                             'reward_std': '0.5346', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '7.544e-05', 'entropy': '0.01716', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '14.94', 'epoch': '0.016'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:30:52]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027515;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027516;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m2\u001b[0m%|▏ | \u001b[1m1\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m00:19<16:16, \u001b[1m19.\u001b[0m92s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m4\u001b[0m%|▍ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m2\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m00:34<13:36, \u001b[1m17.\u001b[0m00s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.002899', 'grad_norm': '1.937', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '9.8e-09', 'num_tokens': '1.326e+05', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '36.66', 'completions/min_length': '20', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '82', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '36.66', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '20', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '82', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.6484', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4793', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9371', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.2445', 'reward': '1.117', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.5346', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '7.544e-05', 'entropy': '0.01716', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '14.94', 'epoch': '0.016'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:31:08] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             4%|▍         | 2/50 [00:34<13:36, 17.00s/it]#015  6%|▌         |                      \n",
+       "                             3/50 [00:50<12:57, 16.55s/it]#015                                                     \n",
+       "                             #015{'loss': '-0.0008521', 'grad_norm': '2.254', 'learning_rate':                     \n",
+       "                             '9.6e-09', 'num_tokens': '2.039e+05', 'completions/mean_length':                      \n",
+       "                             '40.09', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '84', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '40.09',                                        \n",
+       "                             'completions/min_terminated_length': '20',                                            \n",
+       "                             'completions/max_terminated_length': '84',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7734',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4203',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9371',                                               \n",
+       "                             'rewards/format_reward/std': '0.2445', 'reward': '1.242',                             \n",
+       "                             'reward_std': '0.4905', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.0001991', 'entropy': '0.02427', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '16', 'epoch': '0.024'}                                                  \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:31:08]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027523;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027524;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m4\u001b[0m%|▍ | \u001b[1m2\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m00:34<13:36, \u001b[1m17.\u001b[0m00s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m6\u001b[0m%|▌ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m3\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m00:50<12:57, \u001b[1m16.\u001b[0m55s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.0008521', 'grad_norm': '2.254', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '9.6e-09', 'num_tokens': '2.039e+05', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '40.09', 'completions/min_length': '20', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '84', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '40.09', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '20', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '84', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.7734', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4203', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9371', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.2445', 'reward': '1.242', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.4905', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.0001991', 'entropy': '0.02427', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '16', 'epoch': '0.024'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:31:24] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             6%|▌         | 3/50 [00:50<12:57, 16.55s/it]#015  8%|▊         |                      \n",
+       "                             4/50 [01:04<11:53, 15.50s/it]#015                                                     \n",
+       "                             #015{'loss': '0.0005852', 'grad_norm': '1.884', 'learning_rate':                      \n",
+       "                             '9.4e-09', 'num_tokens': '2.644e+05', 'completions/mean_length':                      \n",
+       "                             '34.04', 'completions/min_length': '22', 'completions/max_length':                    \n",
+       "                             '57', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '34.04',                                        \n",
+       "                             'completions/min_terminated_length': '22',                                            \n",
+       "                             'completions/max_terminated_length': '57',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.8281',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.3788',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9214',                                               \n",
+       "                             'rewards/format_reward/std': '0.2711', 'reward': '1.289',                             \n",
+       "                             'reward_std': '0.4769', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.0001854', 'entropy': '0.01301', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '13.88', 'epoch': '0.032'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:31:24]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027531;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027532;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m6\u001b[0m%|▌ | \u001b[1m3\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m00:50<12:57, \u001b[1m16.\u001b[0m55s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m8\u001b[0m%|▊ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m4\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:04<11:53, \u001b[1m15.\u001b[0m50s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.0005852', 'grad_norm': '1.884', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '9.4e-09', 'num_tokens': '2.644e+05', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '34.04', 'completions/min_length': '22', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '57', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '34.04', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '22', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '57', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.8281', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.3788', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9214', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.2711', 'reward': '1.289', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.4769', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.0001854', 'entropy': '0.01301', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '13.88', 'epoch': '0.032'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:31:34] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             8%|▊         | 4/50 [01:04<11:53, 15.50s/it]#015 10%|█         |                      \n",
+       "                             5/50 [01:18<11:01, 14.69s/it]#015                                                     \n",
+       "                             #015{'loss': '-0.004746', 'grad_norm': '2.611', 'learning_rate':                      \n",
+       "                             '9.2e-09', 'num_tokens': '3.378e+05', 'completions/mean_length':                      \n",
+       "                             '35.79', 'completions/min_length': '24', 'completions/max_length':                    \n",
+       "                             '55', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '35.79',                                        \n",
+       "                             'completions/min_terminated_length': '24',                                            \n",
+       "                             'completions/max_terminated_length': '55',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7734',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4203',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9371',                                               \n",
+       "                             'rewards/format_reward/std': '0.2444', 'reward': '1.242',                             \n",
+       "                             'reward_std': '0.4905', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '8.713e-05', 'entropy': '0.01761', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '13.24', 'epoch': '0.04'}                                                \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:31:34]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027539;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027540;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m8\u001b[0m%|▊ | \u001b[1m4\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:04<11:53, \u001b[1m15.\u001b[0m50s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m10\u001b[0m%|█ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m5\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:18<11:01, \u001b[1m14.\u001b[0m69s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.004746', 'grad_norm': '2.611', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '9.2e-09', 'num_tokens': '3.378e+05', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '35.79', 'completions/min_length': '24', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '55', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '35.79', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '24', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '55', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.7734', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4203', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9371', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.2444', 'reward': '1.242', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.4905', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '8.713e-05', 'entropy': '0.01761', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '13.24', 'epoch': '0.04'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:31:45] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             10%|█         | 5/50 [01:18<11:01, 14.69s/it]#015 12%|█▏        |                     \n",
+       "                             6/50 [01:30<10:14, 13.97s/it]#015                                                     \n",
+       "                             #015{'loss': '2.203e-07', 'grad_norm': '1.855', 'learning_rate':                      \n",
+       "                             '9e-09', 'num_tokens': '4.018e+05', 'completions/mean_length':                        \n",
+       "                             '28.62', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '44', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '28.62',                                        \n",
+       "                             'completions/min_terminated_length': '20',                                            \n",
+       "                             'completions/max_terminated_length': '44',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.8516',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.3569',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9136',                                               \n",
+       "                             'rewards/format_reward/std': '0.2828', 'reward': '1.308',                             \n",
+       "                             'reward_std': '0.4707', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '-2.709e-06', 'entropy': '0.01278', 'clip_ratio/low_mean': '0',                       \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '12.56', 'epoch': '0.048'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:31:45]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027547;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027548;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m10\u001b[0m%|█ | \u001b[1m5\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:18<11:01, \u001b[1m14.\u001b[0m69s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m12\u001b[0m%|█▏ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m6\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:30<10:14, \u001b[1m13.\u001b[0m97s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '2.203e-07', 'grad_norm': '1.855', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '9e-09', 'num_tokens': '4.018e+05', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '28.62', 'completions/min_length': '20', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '44', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '28.62', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '20', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '44', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.8516', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.3569', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9136', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.2828', 'reward': '1.308', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.4707', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '-2.709e-06', 'entropy': '0.01278', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '12.56', 'epoch': '0.048'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:32:01] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             12%|█▏        | 6/50 [01:30<10:14, 13.97s/it]#015 14%|█▍        |                     \n",
+       "                             7/50 [01:43<09:49, 13.71s/it]#015                                                     \n",
+       "                             #015{'loss': '-0.000197', 'grad_norm': '5.104', 'learning_rate':                      \n",
+       "                             '8.8e-09', 'num_tokens': '4.587e+05', 'completions/mean_length':                      \n",
+       "                             '27.25', 'completions/min_length': '21', 'completions/max_length':                    \n",
+       "                             '44', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '27.25',                                        \n",
+       "                             'completions/min_terminated_length': '21',                                            \n",
+       "                             'completions/max_terminated_length': '44',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.75',                                              \n",
+       "                             'rewards/tool_call_reward/std': '0.4347',                                             \n",
+       "                             'rewards/format_reward/mean': '0.8979',                                               \n",
+       "                             'rewards/format_reward/std': '0.3049', 'reward': '1.199',                             \n",
+       "                             'reward_std': '0.5379', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.0001013', 'entropy': '0.01624', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '13.17', 'epoch': '0.056'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:32:01]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027555;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027556;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m12\u001b[0m%|█▏ | \u001b[1m6\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:30<10:14, \u001b[1m13.\u001b[0m97s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m14\u001b[0m%|█▍ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m7\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:43<09:49, \u001b[1m13.\u001b[0m71s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.000197', 'grad_norm': '5.104', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '8.8e-09', 'num_tokens': '4.587e+05', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '27.25', 'completions/min_length': '21', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '44', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '27.25', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '21', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '44', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.75', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4347', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.8979', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.3049', 'reward': '1.199', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.5379', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.0001013', 'entropy': '0.01624', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '13.17', 'epoch': '0.056'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:32:18] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             14%|█▍        | 7/50 [01:43<09:49, 13.71s/it]#015 16%|█▌        |                     \n",
+       "                             8/50 [01:58<09:50, 14.06s/it]#015                                                     \n",
+       "                             #015{'loss': '0.005951', 'grad_norm': '1.704', 'learning_rate':                       \n",
+       "                             '8.6e-09', 'num_tokens': '5.259e+05', 'completions/mean_length':                      \n",
+       "                             '38.98', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '109', 'completions/clipped_ratio': '0',                                              \n",
+       "                             'completions/mean_terminated_length': '38.98',                                        \n",
+       "                             'completions/min_terminated_length': '20',                                            \n",
+       "                             'completions/max_terminated_length': '109',                                           \n",
+       "                             'rewards/tool_call_reward/mean': '0.6797',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4684',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9607',                                               \n",
+       "                             'rewards/format_reward/std': '0.1959', 'reward': '1.16',                              \n",
+       "                             'reward_std': '0.5059', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '8.511e-05', 'entropy': '0.01835', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '14.8', 'epoch': '0.064'}                                                \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:32:18]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027563;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027564;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m14\u001b[0m%|█▍ | \u001b[1m7\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:43<09:49, \u001b[1m13.\u001b[0m71s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m16\u001b[0m%|█▌ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m8\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:58<09:50, \u001b[1m14.\u001b[0m06s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.005951', 'grad_norm': '1.704', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '8.6e-09', 'num_tokens': '5.259e+05', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '38.98', 'completions/min_length': '20', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '109', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '38.98', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '20', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '109', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.6797', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4684', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9607', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.1959', 'reward': '1.16', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.5059', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '8.511e-05', 'entropy': '0.01835', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '14.8', 'epoch': '0.064'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:32:29] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             16%|█▌        | 8/50 [01:58<09:50, 14.06s/it]#015 18%|█▊        |                     \n",
+       "                             9/50 [02:10<09:05, 13.31s/it]#015                                                     \n",
+       "                             #015{'loss': '0.001581', 'grad_norm': '2.503', 'learning_rate':                       \n",
+       "                             '8.4e-09', 'num_tokens': '5.847e+05', 'completions/mean_length':                      \n",
+       "                             '32.96', 'completions/min_length': '24', 'completions/max_length':                    \n",
+       "                             '51', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '32.96',                                        \n",
+       "                             'completions/min_terminated_length': '24',                                            \n",
+       "                             'completions/max_terminated_length': '51',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7891',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4096',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9136',                                               \n",
+       "                             'rewards/format_reward/std': '0.2828', 'reward': '1.246',                             \n",
+       "                             'reward_std': '0.5064', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '7.845e-06', 'entropy': '0.01171', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '11.65', 'epoch': '0.072'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:32:29]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027571;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027572;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m16\u001b[0m%|█▌ | \u001b[1m8\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:58<09:50, \u001b[1m14.\u001b[0m06s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m18\u001b[0m%|█▊ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m9\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:10<09:05, \u001b[1m13.\u001b[0m31s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.001581', 'grad_norm': '2.503', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '8.4e-09', 'num_tokens': '5.847e+05', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '32.96', 'completions/min_length': '24', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '51', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '32.96', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '24', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '51', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.7891', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4096', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9136', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.2828', 'reward': '1.246', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.5064', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '7.845e-06', 'entropy': '0.01171', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '11.65', 'epoch': '0.072'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:32:45] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             18%|█▊        | 9/50 [02:10<09:05, 13.31s/it]#015 20%|██        |                     \n",
+       "                             10/50 [02:25<09:19, 13.99s/it]#015                                                    \n",
+       "                             #015{'loss': '-0.003787', 'grad_norm': '3.198', 'learning_rate':                      \n",
+       "                             '8.2e-09', 'num_tokens': '6.406e+05', 'completions/mean_length':                      \n",
+       "                             '30.45', 'completions/min_length': '19', 'completions/max_length':                    \n",
+       "                             '56', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '30.45',                                        \n",
+       "                             'completions/min_terminated_length': '19',                                            \n",
+       "                             'completions/max_terminated_length': '56',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7188',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4514',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9293',                                               \n",
+       "                             'rewards/format_reward/std': '0.2581', 'reward': '1.183',                             \n",
+       "                             'reward_std': '0.5212', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '-1.742e-05', 'entropy': '0.0144', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '15.51', 'epoch': '0.08'}                                                \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:32:45]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027579;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027580;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m18\u001b[0m%|█▊ | \u001b[1m9\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:10<09:05, \u001b[1m13.\u001b[0m31s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m20\u001b[0m%|██ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m10\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:25<09:19, \u001b[1m13.\u001b[0m99s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.003787', 'grad_norm': '3.198', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '8.2e-09', 'num_tokens': '6.406e+05', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '30.45', 'completions/min_length': '19', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '56', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '30.45', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '19', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '56', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.7188', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4514', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9293', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.2581', 'reward': '1.183', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.5212', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '-1.742e-05', 'entropy': '0.0144', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '15.51', 'epoch': '0.08'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:32:55] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             20%|██        | 10/50 [02:25<09:19, 13.99s/it]#015 22%|██▏       |                    \n",
+       "                             11/50 [02:37<08:38, 13.30s/it]#015                                                    \n",
+       "                             #015{'loss': '0.001828', 'grad_norm': '1.217', 'learning_rate':                       \n",
+       "                             '8e-09', 'num_tokens': '6.995e+05', 'completions/mean_length':                        \n",
+       "                             '31.66', 'completions/min_length': '23', 'completions/max_length':                    \n",
+       "                             '47', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '31.66',                                        \n",
+       "                             'completions/min_terminated_length': '23',                                            \n",
+       "                             'completions/max_terminated_length': '47',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.8047',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.398',                                              \n",
+       "                             'rewards/format_reward/mean': '0.9372',                                               \n",
+       "                             'rewards/format_reward/std': '0.2442', 'reward': '1.273',                             \n",
+       "                             'reward_std': '0.4736', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '3.692e-05', 'entropy': '0.01143', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '11.71', 'epoch': '0.088'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:32:55]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027587;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027588;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m20\u001b[0m%|██ | \u001b[1m10\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:25<09:19, \u001b[1m13.\u001b[0m99s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m22\u001b[0m%|██▏ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m11\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:37<08:38, \u001b[1m13.\u001b[0m30s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.001828', 'grad_norm': '1.217', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '8e-09', 'num_tokens': '6.995e+05', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '31.66', 'completions/min_length': '23', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '47', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '31.66', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '23', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '47', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.8047', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.398', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9372', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.2442', 'reward': '1.273', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.4736', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '3.692e-05', 'entropy': '0.01143', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '11.71', 'epoch': '0.088'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:33:06] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             22%|██▏       | 11/50 [02:37<08:38, 13.30s/it]#015 24%|██▍       |                    \n",
+       "                             12/50 [02:48<08:03, 12.73s/it]#015                                                    \n",
+       "                             #015{'loss': '-0.003021', 'grad_norm': '2.149', 'learning_rate':                      \n",
+       "                             '7.8e-09', 'num_tokens': '7.648e+05', 'completions/mean_length':                      \n",
+       "                             '31.52', 'completions/min_length': '22', 'completions/max_length':                    \n",
+       "                             '40', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '31.52',                                        \n",
+       "                             'completions/min_terminated_length': '22',                                            \n",
+       "                             'completions/max_terminated_length': '40',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7266',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4475',                                             \n",
+       "                             'rewards/format_reward/mean': '0.8979',                                               \n",
+       "                             'rewards/format_reward/std': '0.305', 'reward': '1.175',                              \n",
+       "                             'reward_std': '0.5461', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.000162', 'entropy': '0.02277', 'clip_ratio/low_mean': '0',                         \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '11.42', 'epoch': '0.096'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:33:06]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027595;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027596;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m22\u001b[0m%|██▏ | \u001b[1m11\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:37<08:38, \u001b[1m13.\u001b[0m30s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m24\u001b[0m%|██▍ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m12\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:48<08:03, \u001b[1m12.\u001b[0m73s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.003021', 'grad_norm': '2.149', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '7.8e-09', 'num_tokens': '7.648e+05', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '31.52', 'completions/min_length': '22', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '40', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '31.52', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '22', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '40', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.7266', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4475', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.8979', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.305', 'reward': '1.175', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.5461', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.000162', 'entropy': '0.02277', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '11.42', 'epoch': '0.096'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:33:23] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             24%|██▍       | 12/50 [02:48<08:03, 12.73s/it]#015 26%|██▌       |                    \n",
+       "                             13/50 [03:04<08:25, 13.67s/it]#015                                                    \n",
+       "                             #015{'loss': '0.001063', 'grad_norm': '1.959', 'learning_rate':                       \n",
+       "                             '7.6e-09', 'num_tokens': '8.236e+05', 'completions/mean_length':                      \n",
+       "                             '32.22', 'completions/min_length': '21', 'completions/max_length':                    \n",
+       "                             '59', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '32.22',                                        \n",
+       "                             'completions/min_terminated_length': '21',                                            \n",
+       "                             'completions/max_terminated_length': '59',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7188',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4514',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9607',                                               \n",
+       "                             'rewards/format_reward/std': '0.1956', 'reward': '1.199',                             \n",
+       "                             'reward_std': '0.4917', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.000176', 'entropy': '0.0192', 'clip_ratio/low_mean': '0',                          \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '15.83', 'epoch': '0.104'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:33:23]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027603;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027604;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m24\u001b[0m%|██▍ | \u001b[1m12\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:48<08:03, \u001b[1m12.\u001b[0m73s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m26\u001b[0m%|██▌ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m13\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:04<08:25, \u001b[1m13.\u001b[0m67s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.001063', 'grad_norm': '1.959', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '7.6e-09', 'num_tokens': '8.236e+05', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '32.22', 'completions/min_length': '21', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '59', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '32.22', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '21', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '59', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.7188', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4514', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9607', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.1956', 'reward': '1.199', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.4917', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.000176', 'entropy': '0.0192', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '15.83', 'epoch': '0.104'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:33:33] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             26%|██▌       | 13/50 [03:04<08:25, 13.67s/it]#015 28%|██▊       |                    \n",
+       "                             14/50 [03:17<08:04, 13.46s/it]#015                                                    \n",
+       "                             #015{'loss': '0.02065', 'grad_norm': '5.698', 'learning_rate':                        \n",
+       "                             '7.4e-09', 'num_tokens': '8.798e+05', 'completions/mean_length':                      \n",
+       "                             '35.79', 'completions/min_length': '15', 'completions/max_length':                    \n",
+       "                             '73', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '35.79',                                        \n",
+       "                             'completions/min_terminated_length': '15',                                            \n",
+       "                             'completions/max_terminated_length': '73',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7969',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4039',                                             \n",
+       "                             'rewards/format_reward/mean': '0.937', 'rewards/format_reward/std':                   \n",
+       "                             '0.245', 'reward': '1.265', 'reward_std': '0.4783',                                   \n",
+       "                             'frac_reward_zero_std': '0', 'kl': '0.0001055', 'entropy':                            \n",
+       "                             '0.02258', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',                     \n",
+       "                             'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',                              \n",
+       "                             'clip_ratio/region_mean': '0', 'step_time': '12.96', 'epoch':                         \n",
+       "                             '0.112'}                                                                              \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:33:33]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027611;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027612;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m26\u001b[0m%|██▌ | \u001b[1m13\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:04<08:25, \u001b[1m13.\u001b[0m67s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m28\u001b[0m%|██▊ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m14\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:17<08:04, \u001b[1m13.\u001b[0m46s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.02065', 'grad_norm': '5.698', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '7.4e-09', 'num_tokens': '8.798e+05', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '35.79', 'completions/min_length': '15', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '73', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '35.79', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '15', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '73', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.7969', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4039', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.937', 'rewards/format_reward/std': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.245', 'reward': '1.265', 'reward_std': '0.4783', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'frac_reward_zero_std': '0', 'kl': '0.0001055', 'entropy': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.02258', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/region_mean': '0', 'step_time': '12.96', 'epoch': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.112'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:33:49] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             28%|██▊       | 14/50 [03:17<08:04, 13.46s/it]#015 30%|███       |                    \n",
+       "                             15/50 [03:33<08:15, 14.15s/it]#015                                                    \n",
+       "                             #015{'loss': '0.003047', 'grad_norm': '1.325', 'learning_rate':                       \n",
+       "                             '7.2e-09', 'num_tokens': '9.451e+05', 'completions/mean_length':                      \n",
+       "                             '42.41', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '128', 'completions/clipped_ratio': '0.0625',                                         \n",
+       "                             'completions/mean_terminated_length': '36.7',                                         \n",
+       "                             'completions/min_terminated_length': '20',                                            \n",
+       "                             'completions/max_terminated_length': '62',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.8359',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.3718',                                             \n",
+       "                             'rewards/format_reward/mean': '0.8505',                                               \n",
+       "                             'rewards/format_reward/std': '0.3594', 'reward': '1.261',                             \n",
+       "                             'reward_std': '0.5445', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '3.676e-05', 'entropy': '0.009485', 'clip_ratio/low_mean': '0',                       \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '15.74', 'epoch': '0.12'}                                                \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:33:49]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027619;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027620;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m28\u001b[0m%|██▊ | \u001b[1m14\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:17<08:04, \u001b[1m13.\u001b[0m46s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m30\u001b[0m%|███ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m15\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:33<08:15, \u001b[1m14.\u001b[0m15s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.003047', 'grad_norm': '1.325', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '7.2e-09', 'num_tokens': '9.451e+05', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '42.41', 'completions/min_length': '20', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '128', 'completions/clipped_ratio': '0.0625', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '36.7', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '20', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '62', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.8359', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.3718', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.8505', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.3594', 'reward': '1.261', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.5445', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '3.676e-05', 'entropy': '0.009485', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '15.74', 'epoch': '0.12'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:34:05] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             30%|███       | 15/50 [03:33<08:15, 14.15s/it]#015 32%|███▏      |                    \n",
+       "                             16/50 [03:49<08:20, 14.72s/it]#015                                                    \n",
+       "                             #015{'loss': '-0.001221', 'grad_norm': '2.042', 'learning_rate':                      \n",
+       "                             '7e-09', 'num_tokens': '1.022e+06', 'completions/mean_length':                        \n",
+       "                             '32.47', 'completions/min_length': '16', 'completions/max_length':                    \n",
+       "                             '57', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '32.47',                                        \n",
+       "                             'completions/min_terminated_length': '16',                                            \n",
+       "                             'completions/max_terminated_length': '57',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7656',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4253',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9057',                                               \n",
+       "                             'rewards/format_reward/std': '0.2944', 'reward': '1.218',                             \n",
+       "                             'reward_std': '0.5247', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '5.935e-05', 'entropy': '0.02726', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '16.01', 'epoch': '0.128'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:34:05]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027627;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027628;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m30\u001b[0m%|███ | \u001b[1m15\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:33<08:15, \u001b[1m14.\u001b[0m15s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m32\u001b[0m%|███▏ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m16\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:49<08:20, \u001b[1m14.\u001b[0m72s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.001221', 'grad_norm': '2.042', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '7e-09', 'num_tokens': '1.022e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '32.47', 'completions/min_length': '16', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '57', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '32.47', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '16', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '57', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.7656', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4253', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9057', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.2944', 'reward': '1.218', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.5247', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '5.935e-05', 'entropy': '0.02726', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '16.01', 'epoch': '0.128'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:34:22] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             32%|███▏      | 16/50 [03:49<08:20, 14.72s/it]#015 34%|███▍      |                    \n",
+       "                             17/50 [04:05<08:14, 14.98s/it]#015                                                    \n",
+       "                             #015{'loss': '0.007587', 'grad_norm': '2.109', 'learning_rate':                       \n",
+       "                             '6.8e-09', 'num_tokens': '1.075e+06', 'completions/mean_length':                      \n",
+       "                             '36.02', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '128', 'completions/clipped_ratio': '0.0625',                                         \n",
+       "                             'completions/mean_terminated_length': '29.88',                                        \n",
+       "                             'completions/min_terminated_length': '20',                                            \n",
+       "                             'completions/max_terminated_length': '53',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.8594',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.349',                                              \n",
+       "                             'rewards/format_reward/mean': '0.8819',                                               \n",
+       "                             'rewards/format_reward/std': '0.3253', 'reward': '1.3',                               \n",
+       "                             'reward_std': '0.5005', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '3.856e-06', 'entropy': '0.0109', 'clip_ratio/low_mean': '0',                         \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '15.59', 'epoch': '0.136'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:34:22]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027635;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027636;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m32\u001b[0m%|███▏ | \u001b[1m16\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:49<08:20, \u001b[1m14.\u001b[0m72s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m34\u001b[0m%|███▍ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m17\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:05<08:14, \u001b[1m14.\u001b[0m98s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.007587', 'grad_norm': '2.109', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '6.8e-09', 'num_tokens': '1.075e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '36.02', 'completions/min_length': '20', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '128', 'completions/clipped_ratio': '0.0625', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '29.88', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '20', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '53', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.8594', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.349', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.8819', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.3253', 'reward': '1.3', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.5005', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '3.856e-06', 'entropy': '0.0109', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '15.59', 'epoch': '0.136'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:34:32] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             34%|███▍      | 17/50 [04:05<08:14, 14.98s/it]#015 36%|███▌      |                    \n",
+       "                             18/50 [04:17<07:35, 14.23s/it]#015                                                    \n",
+       "                             #015{'loss': '0.004968', 'grad_norm': '2.632', 'learning_rate':                       \n",
+       "                             '6.6e-09', 'num_tokens': '1.14e+06', 'completions/mean_length':                       \n",
+       "                             '33.13', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '65', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '33.13',                                        \n",
+       "                             'completions/min_terminated_length': '20',                                            \n",
+       "                             'completions/max_terminated_length': '65',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.8125',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.3918',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9293',                                               \n",
+       "                             'rewards/format_reward/std': '0.2581', 'reward': '1.277',                             \n",
+       "                             'reward_std': '0.4776', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '8.186e-05', 'entropy': '0.01777', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '12.48', 'epoch': '0.144'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:34:32]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027643;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027644;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m34\u001b[0m%|███▍ | \u001b[1m17\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:05<08:14, \u001b[1m14.\u001b[0m98s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m36\u001b[0m%|███▌ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m18\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:17<07:35, \u001b[1m14.\u001b[0m23s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.004968', 'grad_norm': '2.632', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '6.6e-09', 'num_tokens': '1.14e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '33.13', 'completions/min_length': '20', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '65', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '33.13', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '20', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '65', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.8125', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.3918', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9293', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.2581', 'reward': '1.277', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.4776', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '8.186e-05', 'entropy': '0.01777', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '12.48', 'epoch': '0.144'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:34:48] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             36%|███▌      | 18/50 [04:17<07:35, 14.23s/it]#015 38%|███▊      |                    \n",
+       "                             19/50 [04:33<07:34, 14.65s/it]#015                                                    \n",
+       "                             #015{'loss': '-0.008127', 'grad_norm': '2.291', 'learning_rate':                      \n",
+       "                             '6.4e-09', 'num_tokens': '1.201e+06', 'completions/mean_length':                      \n",
+       "                             '33.46', 'completions/min_length': '16', 'completions/max_length':                    \n",
+       "                             '50', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '33.46',                                        \n",
+       "                             'completions/min_terminated_length': '16',                                            \n",
+       "                             'completions/max_terminated_length': '50',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.8516',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.3569',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9529',                                               \n",
+       "                             'rewards/format_reward/std': '0.2134', 'reward': '1.328',                             \n",
+       "                             'reward_std': '0.4234', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '5.495e-05', 'entropy': '0.01459', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '15.6', 'epoch': '0.152'}                                                \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:34:48]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027651;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027652;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m36\u001b[0m%|███▌ | \u001b[1m18\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:17<07:35, \u001b[1m14.\u001b[0m23s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m38\u001b[0m%|███▊ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m19\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:33<07:34, \u001b[1m14.\u001b[0m65s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.008127', 'grad_norm': '2.291', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '6.4e-09', 'num_tokens': '1.201e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '33.46', 'completions/min_length': '16', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '50', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '33.46', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '16', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '50', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.8516', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.3569', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9529', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.2134', 'reward': '1.328', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.4234', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '5.495e-05', 'entropy': '0.01459', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '15.6', 'epoch': '0.152'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:34:59] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             38%|███▊      | 19/50 [04:33<07:34, 14.65s/it]#015 40%|████      |                    \n",
+       "                             20/50 [04:45<06:55, 13.86s/it]#015                                                    \n",
+       "                             #015{'loss': '-0.002019', 'grad_norm': '2.159', 'learning_rate':                      \n",
+       "                             '6.2e-09', 'num_tokens': '1.272e+06', 'completions/mean_length':                      \n",
+       "                             '32.3', 'completions/min_length': '19', 'completions/max_length':                     \n",
+       "                             '53', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '32.3',                                         \n",
+       "                             'completions/min_terminated_length': '19',                                            \n",
+       "                             'completions/max_terminated_length': '53',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7266',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4475',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9372',                                               \n",
+       "                             'rewards/format_reward/std': '0.2443', 'reward': '1.195',                             \n",
+       "                             'reward_std': '0.511', 'frac_reward_zero_std': '0', 'kl':                             \n",
+       "                             '0.0001452', 'entropy': '0.03184', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '12.01', 'epoch': '0.16'}                                                \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:34:59]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027659;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027660;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m38\u001b[0m%|███▊ | \u001b[1m19\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:33<07:34, \u001b[1m14.\u001b[0m65s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m40\u001b[0m%|████ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m20\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:45<06:55, \u001b[1m13.\u001b[0m86s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.002019', 'grad_norm': '2.159', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '6.2e-09', 'num_tokens': '1.272e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '32.3', 'completions/min_length': '19', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '53', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '32.3', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '19', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '53', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.7266', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4475', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9372', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.2443', 'reward': '1.195', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.511', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.0001452', 'entropy': '0.03184', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '12.01', 'epoch': '0.16'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:35:15] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             40%|████      | 20/50 [04:45<06:55, 13.86s/it]#015 42%|████▏     |                    \n",
+       "                             21/50 [04:57<06:25, 13.28s/it]#015                                                    \n",
+       "                             #015{'loss': '-0.001169', 'grad_norm': '2.646', 'learning_rate':                      \n",
+       "                             '6e-09', 'num_tokens': '1.329e+06', 'completions/mean_length':                        \n",
+       "                             '30.85', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '54', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '30.85',                                        \n",
+       "                             'completions/min_terminated_length': '20',                                            \n",
+       "                             'completions/max_terminated_length': '54',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7969',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4039',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9058',                                               \n",
+       "                             'rewards/format_reward/std': '0.2941', 'reward': '1.25',                              \n",
+       "                             'reward_std': '0.5103', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.0001261', 'entropy': '0.01297', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '11.92', 'epoch': '0.168'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:35:15]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027667;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027668;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m40\u001b[0m%|████ | \u001b[1m20\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:45<06:55, \u001b[1m13.\u001b[0m86s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m42\u001b[0m%|████▏ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m21\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:57<06:25, \u001b[1m13.\u001b[0m28s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.001169', 'grad_norm': '2.646', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '6e-09', 'num_tokens': '1.329e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '30.85', 'completions/min_length': '20', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '54', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '30.85', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '20', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '54', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.7969', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4039', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9058', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.2941', 'reward': '1.25', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.5103', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.0001261', 'entropy': '0.01297', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '11.92', 'epoch': '0.168'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:35:26] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             42%|████▏     | 21/50 [04:57<06:25, 13.28s/it]#015 44%|████▍     |                    \n",
+       "                             22/50 [05:10<06:14, 13.38s/it]#015                                                    \n",
+       "                             #015{'loss': '0.003743', 'grad_norm': '1.94', 'learning_rate':                        \n",
+       "                             '5.8e-09', 'num_tokens': '1.393e+06', 'completions/mean_length':                      \n",
+       "                             '30.69', 'completions/min_length': '18', 'completions/max_length':                    \n",
+       "                             '45', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '30.69',                                        \n",
+       "                             'completions/min_terminated_length': '18',                                            \n",
+       "                             'completions/max_terminated_length': '45',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7109',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4551',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9607',                                               \n",
+       "                             'rewards/format_reward/std': '0.1956', 'reward': '1.191',                             \n",
+       "                             'reward_std': '0.4948', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '9.293e-05', 'entropy': '0.01876', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '13.59', 'epoch': '0.176'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:35:26]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027675;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027676;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m42\u001b[0m%|████▏ | \u001b[1m21\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:57<06:25, \u001b[1m13.\u001b[0m28s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m44\u001b[0m%|████▍ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m22\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:10<06:14, \u001b[1m13.\u001b[0m38s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.003743', 'grad_norm': '1.94', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '5.8e-09', 'num_tokens': '1.393e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '30.69', 'completions/min_length': '18', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '45', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '30.69', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '18', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '45', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.7109', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4551', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9607', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.1956', 'reward': '1.191', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.4948', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '9.293e-05', 'entropy': '0.01876', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '13.59', 'epoch': '0.176'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:35:42] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             44%|████▍     | 22/50 [05:10<06:14, 13.38s/it]#015 46%|████▌     |                    \n",
+       "                             23/50 [05:26<06:23, 14.21s/it]#015                                                    \n",
+       "                             #015{'loss': '0.001856', 'grad_norm': '1.338', 'learning_rate':                       \n",
+       "                             '5.6e-09', 'num_tokens': '1.461e+06', 'completions/mean_length':                      \n",
+       "                             '38.4', 'completions/min_length': '17', 'completions/max_length':                     \n",
+       "                             '128', 'completions/clipped_ratio': '0.0625',                                         \n",
+       "                             'completions/mean_terminated_length': '32.43',                                        \n",
+       "                             'completions/min_terminated_length': '17',                                            \n",
+       "                             'completions/max_terminated_length': '47',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.8359',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.3718',                                             \n",
+       "                             'rewards/format_reward/mean': '0.882', 'rewards/format_reward/std':                   \n",
+       "                             '0.3252', 'reward': '1.277', 'reward_std': '0.5139',                                  \n",
+       "                             'frac_reward_zero_std': '0', 'kl': '7.992e-06', 'entropy':                            \n",
+       "                             '0.01434', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',                     \n",
+       "                             'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',                              \n",
+       "                             'clip_ratio/region_mean': '0', 'step_time': '16.14', 'epoch':                         \n",
+       "                             '0.184'}                                                                              \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:35:42]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027683;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027684;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m44\u001b[0m%|████▍ | \u001b[1m22\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:10<06:14, \u001b[1m13.\u001b[0m38s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m46\u001b[0m%|████▌ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m23\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:26<06:23, \u001b[1m14.\u001b[0m21s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.001856', 'grad_norm': '1.338', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '5.6e-09', 'num_tokens': '1.461e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '38.4', 'completions/min_length': '17', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '128', 'completions/clipped_ratio': '0.0625', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '32.43', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '17', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '47', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.8359', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.3718', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.882', 'rewards/format_reward/std': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.3252', 'reward': '1.277', 'reward_std': '0.5139', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'frac_reward_zero_std': '0', 'kl': '7.992e-06', 'entropy': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.01434', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/region_mean': '0', 'step_time': '16.14', 'epoch': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.184'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:35:58] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             46%|████▌     | 23/50 [05:26<06:23, 14.21s/it]#015 48%|████▊     |                    \n",
+       "                             24/50 [05:40<06:02, 13.94s/it]#015                                                    \n",
+       "                             #015{'loss': '0.001372', 'grad_norm': '1.178', 'learning_rate':                       \n",
+       "                             '5.4e-09', 'num_tokens': '1.526e+06', 'completions/mean_length':                      \n",
+       "                             '39.73', 'completions/min_length': '21', 'completions/max_length':                    \n",
+       "                             '84', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '39.73',                                        \n",
+       "                             'completions/min_terminated_length': '21',                                            \n",
+       "                             'completions/max_terminated_length': '84',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.5547',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.499',                                              \n",
+       "                             'rewards/format_reward/mean': '0.9214',                                               \n",
+       "                             'rewards/format_reward/std': '0.2711', 'reward': '1.015',                             \n",
+       "                             'reward_std': '0.5579', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.0002224', 'entropy': '0.06501', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '13.29', 'epoch': '0.192'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:35:58]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027691;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027692;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m46\u001b[0m%|████▌ | \u001b[1m23\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:26<06:23, \u001b[1m14.\u001b[0m21s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m48\u001b[0m%|████▊ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m24\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:40<06:02, \u001b[1m13.\u001b[0m94s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.001372', 'grad_norm': '1.178', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '5.4e-09', 'num_tokens': '1.526e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '39.73', 'completions/min_length': '21', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '84', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '39.73', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '21', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '84', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.5547', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.499', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9214', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.2711', 'reward': '1.015', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.5579', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.0002224', 'entropy': '0.06501', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '13.29', 'epoch': '0.192'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:36:14] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             48%|████▊     | 24/50 [05:40<06:02, 13.94s/it]#015 50%|█████     |                    \n",
+       "                             25/50 [05:54<05:53, 14.16s/it]#015                                                    \n",
+       "                             #015{'loss': '0.002559', 'grad_norm': '2.119', 'learning_rate':                       \n",
+       "                             '5.2e-09', 'num_tokens': '1.586e+06', 'completions/mean_length':                      \n",
+       "                             '36.06', 'completions/min_length': '19', 'completions/max_length':                    \n",
+       "                             '56', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '36.06',                                        \n",
+       "                             'completions/min_terminated_length': '19',                                            \n",
+       "                             'completions/max_terminated_length': '56',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.8047',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.398',                                              \n",
+       "                             'rewards/format_reward/mean': '0.9136',                                               \n",
+       "                             'rewards/format_reward/std': '0.2829', 'reward': '1.261',                             \n",
+       "                             'reward_std': '0.4985', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '6.807e-05', 'entropy': '0.01488', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '14.66', 'epoch': '0.2'}                                                 \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:36:14]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027699;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027700;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m48\u001b[0m%|████▊ | \u001b[1m24\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:40<06:02, \u001b[1m13.\u001b[0m94s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m50\u001b[0m%|█████ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m25\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:54<05:53, \u001b[1m14.\u001b[0m16s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.002559', 'grad_norm': '2.119', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '5.2e-09', 'num_tokens': '1.586e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '36.06', 'completions/min_length': '19', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '56', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '36.06', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '19', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '56', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.8047', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.398', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9136', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.2829', 'reward': '1.261', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.4985', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '6.807e-05', 'entropy': '0.01488', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '14.66', 'epoch': '0.2'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:36:25] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             50%|█████     | 25/50 [05:54<05:53, 14.16s/it]#015 52%|█████▏    |                    \n",
+       "                             26/50 [06:08<05:35, 14.00s/it]#015                                                    \n",
+       "                             #015{'loss': '0.003904', 'grad_norm': '3.429', 'learning_rate':                       \n",
+       "                             '5e-09', 'num_tokens': '1.65e+06', 'completions/mean_length':                         \n",
+       "                             '32.04', 'completions/min_length': '21', 'completions/max_length':                    \n",
+       "                             '81', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '32.04',                                        \n",
+       "                             'completions/min_terminated_length': '21',                                            \n",
+       "                             'completions/max_terminated_length': '81',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7109',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4551',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9292',                                               \n",
+       "                             'rewards/format_reward/std': '0.2583', 'reward': '1.176',                             \n",
+       "                             'reward_std': '0.5239', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '7.857e-05', 'entropy': '0.0258', 'clip_ratio/low_mean': '0',                         \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '13.61', 'epoch': '0.208'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:36:25]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027707;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027708;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m50\u001b[0m%|█████ | \u001b[1m25\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:54<05:53, \u001b[1m14.\u001b[0m16s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m52\u001b[0m%|█████▏ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m26\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:08<05:35, \u001b[1m14.\u001b[0m00s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.003904', 'grad_norm': '3.429', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '5e-09', 'num_tokens': '1.65e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '32.04', 'completions/min_length': '21', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '81', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '32.04', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '21', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '81', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.7109', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4551', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9292', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.2583', 'reward': '1.176', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.5239', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '7.857e-05', 'entropy': '0.0258', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '13.61', 'epoch': '0.208'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:36:41] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             52%|█████▏    | 26/50 [06:08<05:35, 14.00s/it]#015 54%|█████▍    |                    \n",
+       "                             27/50 [06:24<05:32, 14.46s/it]#015                                                    \n",
+       "                             #015{'loss': '0.04762', 'grad_norm': '1.92', 'learning_rate':                         \n",
+       "                             '4.8e-09', 'num_tokens': '1.719e+06', 'completions/mean_length':                      \n",
+       "                             '38.88', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '128', 'completions/clipped_ratio': '0.04688',                                        \n",
+       "                             'completions/mean_terminated_length': '34.49',                                        \n",
+       "                             'completions/min_terminated_length': '20',                                            \n",
+       "                             'completions/max_terminated_length': '59',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.6953',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4621',                                             \n",
+       "                             'rewards/format_reward/mean': '0.8661',                                               \n",
+       "                             'rewards/format_reward/std': '0.3434', 'reward': '1.128',                             \n",
+       "                             'reward_std': '0.5804', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '5.935e-05', 'entropy': '0.01181', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '15.51', 'epoch': '0.216'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:36:41]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027715;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027716;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m52\u001b[0m%|█████▏ | \u001b[1m26\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:08<05:35, \u001b[1m14.\u001b[0m00s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m54\u001b[0m%|█████▍ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m27\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:24<05:32, \u001b[1m14.\u001b[0m46s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.04762', 'grad_norm': '1.92', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '4.8e-09', 'num_tokens': '1.719e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '38.88', 'completions/min_length': '20', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '128', 'completions/clipped_ratio': '0.04688', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '34.49', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '20', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '59', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.6953', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4621', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.8661', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.3434', 'reward': '1.128', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.5804', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '5.935e-05', 'entropy': '0.01181', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '15.51', 'epoch': '0.216'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:36:52] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             54%|█████▍    | 27/50 [06:24<05:32, 14.46s/it]#015 56%|█████▌    |                    \n",
+       "                             28/50 [06:38<05:15, 14.35s/it]#015                                                    \n",
+       "                             #015{'loss': '-0.004456', 'grad_norm': '1.489', 'learning_rate':                      \n",
+       "                             '4.6e-09', 'num_tokens': '1.786e+06', 'completions/mean_length':                      \n",
+       "                             '33.55', 'completions/min_length': '19', 'completions/max_length':                    \n",
+       "                             '59', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '33.55',                                        \n",
+       "                             'completions/min_terminated_length': '19',                                            \n",
+       "                             'completions/max_terminated_length': '59',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.6875',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4653',                                             \n",
+       "                             'rewards/format_reward/mean': '0.8979',                                               \n",
+       "                             'rewards/format_reward/std': '0.3049', 'reward': '1.136',                             \n",
+       "                             'reward_std': '0.5572', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.000104', 'entropy': '0.01498', 'clip_ratio/low_mean': '0',                         \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '14.11', 'epoch': '0.224'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:36:52]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027723;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027724;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m54\u001b[0m%|█████▍ | \u001b[1m27\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:24<05:32, \u001b[1m14.\u001b[0m46s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m56\u001b[0m%|█████▌ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m28\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:38<05:15, \u001b[1m14.\u001b[0m35s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.004456', 'grad_norm': '1.489', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '4.6e-09', 'num_tokens': '1.786e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '33.55', 'completions/min_length': '19', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '59', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '33.55', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '19', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '59', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.6875', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4653', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.8979', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.3049', 'reward': '1.136', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.5572', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.000104', 'entropy': '0.01498', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '14.11', 'epoch': '0.224'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:37:08] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             56%|█████▌    | 28/50 [06:38<05:15, 14.35s/it]#015 58%|█████▊    |                    \n",
+       "                             29/50 [06:50<04:50, 13.85s/it]#015                                                    \n",
+       "                             #015{'loss': '0.002348', 'grad_norm': '2.229', 'learning_rate':                       \n",
+       "                             '4.4e-09', 'num_tokens': '1.85e+06', 'completions/mean_length':                       \n",
+       "                             '29.97', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '46', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '29.97',                                        \n",
+       "                             'completions/min_terminated_length': '20',                                            \n",
+       "                             'completions/max_terminated_length': '46',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7891',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4096',                                             \n",
+       "                             'rewards/format_reward/mean': '0.945', 'rewards/format_reward/std':                   \n",
+       "                             '0.2294', 'reward': '1.262', 'reward_std': '0.4739',                                  \n",
+       "                             'frac_reward_zero_std': '0', 'kl': '0.000111', 'entropy':                             \n",
+       "                             '0.03362', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',                     \n",
+       "                             'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',                              \n",
+       "                             'clip_ratio/region_mean': '0', 'step_time': '12.65', 'epoch':                         \n",
+       "                             '0.232'}                                                                              \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:37:08]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027731;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027732;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m56\u001b[0m%|█████▌ | \u001b[1m28\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:38<05:15, \u001b[1m14.\u001b[0m35s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m58\u001b[0m%|█████▊ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m29\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:50<04:50, \u001b[1m13.\u001b[0m85s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.002348', 'grad_norm': '2.229', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '4.4e-09', 'num_tokens': '1.85e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '29.97', 'completions/min_length': '20', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '46', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '29.97', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '20', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '46', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.7891', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4096', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.945', 'rewards/format_reward/std': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.2294', 'reward': '1.262', 'reward_std': '0.4739', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'frac_reward_zero_std': '0', 'kl': '0.000111', 'entropy': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.03362', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/region_mean': '0', 'step_time': '12.65', 'epoch': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.232'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:37:24] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             58%|█████▊    | 29/50 [06:50<04:50, 13.85s/it]#015 60%|██████    |                    \n",
+       "                             30/50 [07:06<04:47, 14.38s/it]#015                                                    \n",
+       "                             #015{'loss': '0.003175', 'grad_norm': '0.7581', 'learning_rate':                      \n",
+       "                             '4.2e-09', 'num_tokens': '1.919e+06', 'completions/mean_length':                      \n",
+       "                             '41.95', 'completions/min_length': '19', 'completions/max_length':                    \n",
+       "                             '128', 'completions/clipped_ratio': '0.0625',                                         \n",
+       "                             'completions/mean_terminated_length': '36.21',                                        \n",
+       "                             'completions/min_terminated_length': '19',                                            \n",
+       "                             'completions/max_terminated_length': '90',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.6953',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4621',                                             \n",
+       "                             'rewards/format_reward/mean': '0.8732',                                               \n",
+       "                             'rewards/format_reward/std': '0.3369', 'reward': '1.132',                             \n",
+       "                             'reward_std': '0.5751', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.0001467', 'entropy': '0.03984', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '15.6', 'epoch': '0.24'}                                                 \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:37:24]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027739;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027740;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m58\u001b[0m%|█████▊ | \u001b[1m29\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:50<04:50, \u001b[1m13.\u001b[0m85s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m60\u001b[0m%|██████ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m30\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:06<04:47, \u001b[1m14.\u001b[0m38s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.003175', 'grad_norm': '0.7581', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '4.2e-09', 'num_tokens': '1.919e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '41.95', 'completions/min_length': '19', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '128', 'completions/clipped_ratio': '0.0625', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '36.21', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '19', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '90', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.6953', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4621', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.8732', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.3369', 'reward': '1.132', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.5751', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.0001467', 'entropy': '0.03984', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '15.6', 'epoch': '0.24'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:37:35] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             60%|██████    | 30/50 [07:06<04:47, 14.38s/it]#015 62%|██████▏   |                    \n",
+       "                             31/50 [07:18<04:20, 13.73s/it]#015                                                    \n",
+       "                             #015{'loss': '-0.005809', 'grad_norm': '2.727', 'learning_rate':                      \n",
+       "                             '4e-09', 'num_tokens': '1.974e+06', 'completions/mean_length':                        \n",
+       "                             '32.5', 'completions/min_length': '21', 'completions/max_length':                     \n",
+       "                             '41', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '32.5',                                         \n",
+       "                             'completions/min_terminated_length': '21',                                            \n",
+       "                             'completions/max_terminated_length': '41',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7734',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4203',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9214',                                               \n",
+       "                             'rewards/format_reward/std': '0.2709', 'reward': '1.234',                             \n",
+       "                             'reward_std': '0.5062', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '8.09e-05', 'entropy': '0.01413', 'clip_ratio/low_mean': '0',                         \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '12.19', 'epoch': '0.248'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:37:35]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027747;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027748;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m60\u001b[0m%|██████ | \u001b[1m30\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:06<04:47, \u001b[1m14.\u001b[0m38s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m62\u001b[0m%|██████▏ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m31\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:18<04:20, \u001b[1m13.\u001b[0m73s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.005809', 'grad_norm': '2.727', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '4e-09', 'num_tokens': '1.974e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '32.5', 'completions/min_length': '21', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '41', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '32.5', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '21', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '41', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.7734', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4203', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9214', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.2709', 'reward': '1.234', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.5062', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '8.09e-05', 'entropy': '0.01413', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '12.19', 'epoch': '0.248'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:37:51] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             62%|██████▏   | 31/50 [07:18<04:20, 13.73s/it]#015 64%|██████▍   |                    \n",
+       "                             32/50 [07:32<04:10, 13.89s/it]#015                                                    \n",
+       "                             #015{'loss': '0.0004126', 'grad_norm': '2.2', 'learning_rate':                        \n",
+       "                             '3.8e-09', 'num_tokens': '2.054e+06', 'completions/mean_length':                      \n",
+       "                             '35.3', 'completions/min_length': '22', 'completions/max_length':                     \n",
+       "                             '62', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '35.3',                                         \n",
+       "                             'completions/min_terminated_length': '22',                                            \n",
+       "                             'completions/max_terminated_length': '62',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7812',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.415',                                              \n",
+       "                             'rewards/format_reward/mean': '0.9371',                                               \n",
+       "                             'rewards/format_reward/std': '0.2446', 'reward': '1.25',                              \n",
+       "                             'reward_std': '0.4866', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '8.267e-05', 'entropy': '0.022', 'clip_ratio/low_mean': '0',                          \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '14.27', 'epoch': '0.256'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:37:51]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027755;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027756;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m62\u001b[0m%|██████▏ | \u001b[1m31\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:18<04:20, \u001b[1m13.\u001b[0m73s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m64\u001b[0m%|██████▍ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m32\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:32<04:10, \u001b[1m13.\u001b[0m89s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.0004126', 'grad_norm': '2.2', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '3.8e-09', 'num_tokens': '2.054e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '35.3', 'completions/min_length': '22', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '62', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '35.3', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '22', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '62', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.7812', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.415', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9371', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.2446', 'reward': '1.25', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.4866', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '8.267e-05', 'entropy': '0.022', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '14.27', 'epoch': '0.256'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:38:01] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             64%|██████▍   | 32/50 [07:32<04:10, 13.89s/it]#015 66%|██████▌   |                    \n",
+       "                             33/50 [07:46<03:55, 13.87s/it]#015                                                    \n",
+       "                             #015{'loss': '0.01859', 'grad_norm': '2.834', 'learning_rate':                        \n",
+       "                             '3.6e-09', 'num_tokens': '2.112e+06', 'completions/mean_length':                      \n",
+       "                             '38.46', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '94', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '38.46',                                        \n",
+       "                             'completions/min_terminated_length': '20',                                            \n",
+       "                             'completions/max_terminated_length': '94',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7734',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4203',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9229',                                               \n",
+       "                             'rewards/format_reward/std': '0.2665', 'reward': '1.235',                             \n",
+       "                             'reward_std': '0.5045', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.0002194', 'entropy': '0.0316', 'clip_ratio/low_mean': '0',                         \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '13.8', 'epoch': '0.264'}                                                \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:38:01]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027763;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027764;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m64\u001b[0m%|██████▍ | \u001b[1m32\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:32<04:10, \u001b[1m13.\u001b[0m89s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m66\u001b[0m%|██████▌ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m33\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:46<03:55, \u001b[1m13.\u001b[0m87s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.01859', 'grad_norm': '2.834', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '3.6e-09', 'num_tokens': '2.112e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '38.46', 'completions/min_length': '20', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '94', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '38.46', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '20', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '94', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.7734', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4203', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9229', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.2665', 'reward': '1.235', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.5045', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.0002194', 'entropy': '0.0316', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '13.8', 'epoch': '0.264'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:38:12] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             66%|██████▌   | 33/50 [07:46<03:55, 13.87s/it]#015 68%|██████▊   |                    \n",
+       "                             34/50 [07:59<03:35, 13.50s/it]#015                                                    \n",
+       "                             #015{'loss': '0.02429', 'grad_norm': '3.956', 'learning_rate':                        \n",
+       "                             '3.4e-09', 'num_tokens': '2.19e+06', 'completions/mean_length':                       \n",
+       "                             '29.49', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '60', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '29.49',                                        \n",
+       "                             'completions/min_terminated_length': '20',                                            \n",
+       "                             'completions/max_terminated_length': '60',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7734',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4203',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9293',                                               \n",
+       "                             'rewards/format_reward/std': '0.2579', 'reward': '1.238',                             \n",
+       "                             'reward_std': '0.4983', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '3.041e-05', 'entropy': '0.022', 'clip_ratio/low_mean': '0',                          \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '12.63', 'epoch': '0.272'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:38:12]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027771;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027772;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m66\u001b[0m%|██████▌ | \u001b[1m33\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:46<03:55, \u001b[1m13.\u001b[0m87s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m68\u001b[0m%|██████▊ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m34\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:59<03:35, \u001b[1m13.\u001b[0m50s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.02429', 'grad_norm': '3.956', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '3.4e-09', 'num_tokens': '2.19e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '29.49', 'completions/min_length': '20', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '60', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '29.49', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '20', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '60', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.7734', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4203', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9293', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.2579', 'reward': '1.238', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.4983', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '3.041e-05', 'entropy': '0.022', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '12.63', 'epoch': '0.272'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:38:29] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             68%|██████▊   | 34/50 [07:59<03:35, 13.50s/it]#015 70%|███████   |                    \n",
+       "                             35/50 [08:14<03:27, 13.83s/it]#015                                                    \n",
+       "                             #015{'loss': '0.002602', 'grad_norm': '2.153', 'learning_rate':                       \n",
+       "                             '3.2e-09', 'num_tokens': '2.246e+06', 'completions/mean_length':                      \n",
+       "                             '30.71', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '57', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '30.71',                                        \n",
+       "                             'completions/min_terminated_length': '20',                                            \n",
+       "                             'completions/max_terminated_length': '57',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.8438',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.3645',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9136',                                               \n",
+       "                             'rewards/format_reward/std': '0.283', 'reward': '1.301',                              \n",
+       "                             'reward_std': '0.4758', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '1.009e-05', 'entropy': '0.0104', 'clip_ratio/low_mean': '0',                         \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '14.59', 'epoch': '0.28'}                                                \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:38:29]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027779;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027780;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m68\u001b[0m%|██████▊ | \u001b[1m34\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:59<03:35, \u001b[1m13.\u001b[0m50s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m70\u001b[0m%|███████ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m35\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m08:14<03:27, \u001b[1m13.\u001b[0m83s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.002602', 'grad_norm': '2.153', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '3.2e-09', 'num_tokens': '2.246e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '30.71', 'completions/min_length': '20', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '57', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '30.71', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '20', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '57', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.8438', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.3645', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9136', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.283', 'reward': '1.301', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.4758', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '1.009e-05', 'entropy': '0.0104', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '14.59', 'epoch': '0.28'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:38:45] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             70%|███████   | 35/50 [08:14<03:27, 13.83s/it]#015 72%|███████▏  |                    \n",
+       "                             36/50 [08:29<03:18, 14.19s/it]#015                                                    \n",
+       "                             #015{'loss': '0.005399', 'grad_norm': '1.961', 'learning_rate':                       \n",
+       "                             '3e-09', 'num_tokens': '2.305e+06', 'completions/mean_length':                        \n",
+       "                             '35.13', 'completions/min_length': '19', 'completions/max_length':                    \n",
+       "                             '119', 'completions/clipped_ratio': '0',                                              \n",
+       "                             'completions/mean_terminated_length': '35.13',                                        \n",
+       "                             'completions/min_terminated_length': '19',                                            \n",
+       "                             'completions/max_terminated_length': '119',                                           \n",
+       "                             'rewards/tool_call_reward/mean': '0.6719',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4714',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9215',                                               \n",
+       "                             'rewards/format_reward/std': '0.2707', 'reward': '1.133',                             \n",
+       "                             'reward_std': '0.5419', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.0001331', 'entropy': '0.01822', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '15', 'epoch': '0.288'}                                                  \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:38:45]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027787;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027788;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m70\u001b[0m%|███████ | \u001b[1m35\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m08:14<03:27, \u001b[1m13.\u001b[0m83s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m72\u001b[0m%|███████▏ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m36\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m08:29<03:18, \u001b[1m14.\u001b[0m19s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.005399', 'grad_norm': '1.961', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '3e-09', 'num_tokens': '2.305e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '35.13', 'completions/min_length': '19', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '119', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '35.13', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '19', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '119', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.6719', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4714', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9215', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.2707', 'reward': '1.133', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.5419', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.0001331', 'entropy': '0.01822', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '15', 'epoch': '0.288'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:39:01] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             72%|███████▏  | 36/50 [08:29<03:18, 14.19s/it]#015 74%|███████▍  |                    \n",
+       "                             37/50 [08:44<03:08, 14.47s/it]#015                                                    \n",
+       "                             #015{'loss': '0.02714', 'grad_norm': '4.003', 'learning_rate':                        \n",
+       "                             '2.8e-09', 'num_tokens': '2.393e+06', 'completions/mean_length':                      \n",
+       "                             '32.17', 'completions/min_length': '22', 'completions/max_length':                    \n",
+       "                             '71', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '32.17',                                        \n",
+       "                             'completions/min_terminated_length': '22',                                            \n",
+       "                             'completions/max_terminated_length': '71',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7656',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4253',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9214',                                               \n",
+       "                             'rewards/format_reward/std': '0.2712', 'reward': '1.226',                             \n",
+       "                             'reward_std': '0.5098', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.0001479', 'entropy': '0.0256', 'clip_ratio/low_mean': '0',                         \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '15.12', 'epoch': '0.296'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:39:01]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027795;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027796;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m72\u001b[0m%|███████▏ | \u001b[1m36\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m08:29<03:18, \u001b[1m14.\u001b[0m19s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m74\u001b[0m%|███████▍ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m37\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m08:44<03:08, \u001b[1m14.\u001b[0m47s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.02714', 'grad_norm': '4.003', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '2.8e-09', 'num_tokens': '2.393e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '32.17', 'completions/min_length': '22', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '71', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '32.17', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '22', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '71', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.7656', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4253', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9214', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.2712', 'reward': '1.226', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.5098', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.0001479', 'entropy': '0.0256', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '15.12', 'epoch': '0.296'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:39:17] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             74%|███████▍  | 37/50 [08:44<03:08, 14.47s/it]#015 76%|███████▌  |                    \n",
+       "                             38/50 [09:00<03:00, 15.04s/it]#015                                                    \n",
+       "                             #015{'loss': '-0.0003246', 'grad_norm': '3.2', 'learning_rate':                       \n",
+       "                             '2.6e-09', 'num_tokens': '2.457e+06', 'completions/mean_length':                      \n",
+       "                             '33.5', 'completions/min_length': '21', 'completions/max_length':                     \n",
+       "                             '62', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '33.5',                                         \n",
+       "                             'completions/min_terminated_length': '21',                                            \n",
+       "                             'completions/max_terminated_length': '62',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7188',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4514',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9151',                                               \n",
+       "                             'rewards/format_reward/std': '0.2784', 'reward': '1.176',                             \n",
+       "                             'reward_std': '0.5335', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '9.85e-05', 'entropy': '0.02256', 'clip_ratio/low_mean': '0',                         \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '16.36', 'epoch': '0.304'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:39:17]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027803;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027804;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m74\u001b[0m%|███████▍ | \u001b[1m37\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m08:44<03:08, \u001b[1m14.\u001b[0m47s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m76\u001b[0m%|███████▌ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m38\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:00<03:00, \u001b[1m15.\u001b[0m04s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.0003246', 'grad_norm': '3.2', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '2.6e-09', 'num_tokens': '2.457e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '33.5', 'completions/min_length': '21', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '62', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '33.5', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '21', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '62', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.7188', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4514', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9151', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.2784', 'reward': '1.176', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.5335', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '9.85e-05', 'entropy': '0.02256', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '16.36', 'epoch': '0.304'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:39:33] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             76%|███████▌  | 38/50 [09:00<03:00, 15.04s/it]#015 78%|███████▊  |                    \n",
+       "                             39/50 [09:16<02:47, 15.26s/it]#015                                                    \n",
+       "                             #015{'loss': '0.006804', 'grad_norm': '1.223', 'learning_rate':                       \n",
+       "                             '2.4e-09', 'num_tokens': '2.522e+06', 'completions/mean_length':                      \n",
+       "                             '42.23', 'completions/min_length': '22', 'completions/max_length':                    \n",
+       "                             '128', 'completions/clipped_ratio': '0.0625',                                         \n",
+       "                             'completions/mean_terminated_length': '36.52',                                        \n",
+       "                             'completions/min_terminated_length': '22',                                            \n",
+       "                             'completions/max_terminated_length': '60',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7734',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4203',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9528',                                               \n",
+       "                             'rewards/format_reward/std': '0.2136', 'reward': '1.25',                              \n",
+       "                             'reward_std': '0.4741', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '2.231e-05', 'entropy': '0.01186', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '15.77', 'epoch': '0.312'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:39:33]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027811;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027812;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m76\u001b[0m%|███████▌ | \u001b[1m38\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:00<03:00, \u001b[1m15.\u001b[0m04s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m78\u001b[0m%|███████▊ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m39\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:16<02:47, \u001b[1m15.\u001b[0m26s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.006804', 'grad_norm': '1.223', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '2.4e-09', 'num_tokens': '2.522e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '42.23', 'completions/min_length': '22', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '128', 'completions/clipped_ratio': '0.0625', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '36.52', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '22', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '60', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.7734', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4203', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9528', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.2136', 'reward': '1.25', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.4741', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '2.231e-05', 'entropy': '0.01186', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '15.77', 'epoch': '0.312'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:39:43] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             78%|███████▊  | 39/50 [09:16<02:47, 15.26s/it]#015 80%|████████  |                    \n",
+       "                             40/50 [09:28<02:23, 14.38s/it]#015                                                    \n",
+       "                             #015{'loss': '0.002718', 'grad_norm': '2.099', 'learning_rate':                       \n",
+       "                             '2.2e-09', 'num_tokens': '2.585e+06', 'completions/mean_length':                      \n",
+       "                             '36.89', 'completions/min_length': '22', 'completions/max_length':                    \n",
+       "                             '59', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '36.89',                                        \n",
+       "                             'completions/min_terminated_length': '22',                                            \n",
+       "                             'completions/max_terminated_length': '59',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7812',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.415',                                              \n",
+       "                             'rewards/format_reward/mean': '0.945', 'rewards/format_reward/std':                   \n",
+       "                             '0.2296', 'reward': '1.254', 'reward_std': '0.4783',                                  \n",
+       "                             'frac_reward_zero_std': '0', 'kl': '0.0003581', 'entropy':                            \n",
+       "                             '0.1239', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',                      \n",
+       "                             'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',                              \n",
+       "                             'clip_ratio/region_mean': '0', 'step_time': '12.31', 'epoch':                         \n",
+       "                             '0.32'}                                                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:39:43]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027819;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027820;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m78\u001b[0m%|███████▊ | \u001b[1m39\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:16<02:47, \u001b[1m15.\u001b[0m26s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m80\u001b[0m%|████████ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m40\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:28<02:23, \u001b[1m14.\u001b[0m38s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.002718', 'grad_norm': '2.099', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '2.2e-09', 'num_tokens': '2.585e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '36.89', 'completions/min_length': '22', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '59', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '36.89', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '22', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '59', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.7812', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.415', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.945', 'rewards/format_reward/std': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.2296', 'reward': '1.254', 'reward_std': '0.4783', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'frac_reward_zero_std': '0', 'kl': '0.0003581', 'entropy': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.1239', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/region_mean': '0', 'step_time': '12.31', 'epoch': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.32'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:40:00] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             80%|████████  | 40/50 [09:28<02:23, 14.38s/it]#015 82%|████████▏ |                    \n",
+       "                             41/50 [09:45<02:15, 15.03s/it]#015                                                    \n",
+       "                             #015{'loss': '0.001328', 'grad_norm': '2.035', 'learning_rate':                       \n",
+       "                             '2e-09', 'num_tokens': '2.647e+06', 'completions/mean_length':                        \n",
+       "                             '31.13', 'completions/min_length': '19', 'completions/max_length':                    \n",
+       "                             '77', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '31.13',                                        \n",
+       "                             'completions/min_terminated_length': '19',                                            \n",
+       "                             'completions/max_terminated_length': '77',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7734',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4203',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9058',                                               \n",
+       "                             'rewards/format_reward/std': '0.2939', 'reward': '1.226',                             \n",
+       "                             'reward_std': '0.5212', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '4.548e-05', 'entropy': '0.016', 'clip_ratio/low_mean': '0',                          \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '16.54', 'epoch': '0.328'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:40:00]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027827;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027828;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m80\u001b[0m%|████████ | \u001b[1m40\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:28<02:23, \u001b[1m14.\u001b[0m38s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m82\u001b[0m%|████████▏ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m41\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:45<02:15, \u001b[1m15.\u001b[0m03s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.001328', 'grad_norm': '2.035', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '2e-09', 'num_tokens': '2.647e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '31.13', 'completions/min_length': '19', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '77', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '31.13', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '19', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '77', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.7734', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4203', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9058', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.2939', 'reward': '1.226', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.5212', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '4.548e-05', 'entropy': '0.016', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '16.54', 'epoch': '0.328'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:40:14] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             82%|████████▏ | 41/50 [09:45<02:15, 15.03s/it]#015 84%|████████▍ |                    \n",
+       "                             42/50 [09:57<01:54, 14.35s/it]#015                                                    \n",
+       "                             #015{'loss': '-0.003071', 'grad_norm': '3.307', 'learning_rate':                      \n",
+       "                             '1.8e-09', 'num_tokens': '2.724e+06', 'completions/mean_length':                      \n",
+       "                             '32.88', 'completions/min_length': '22', 'completions/max_length':                    \n",
+       "                             '45', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '32.88',                                        \n",
+       "                             'completions/min_terminated_length': '22',                                            \n",
+       "                             'completions/max_terminated_length': '45',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.8516',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.3569',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9607',                                               \n",
+       "                             'rewards/format_reward/std': '0.1957', 'reward': '1.332',                             \n",
+       "                             'reward_std': '0.4132', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.000215', 'entropy': '0.02876', 'clip_ratio/low_mean': '0',                         \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '12.75', 'epoch': '0.336'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:40:14]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027835;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027836;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m82\u001b[0m%|████████▏ | \u001b[1m41\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:45<02:15, \u001b[1m15.\u001b[0m03s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m84\u001b[0m%|████████▍ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m42\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:57<01:54, \u001b[1m14.\u001b[0m35s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.003071', 'grad_norm': '3.307', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '1.8e-09', 'num_tokens': '2.724e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '32.88', 'completions/min_length': '22', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '45', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '32.88', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '22', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '45', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.8516', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.3569', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9607', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.1957', 'reward': '1.332', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.4132', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.000215', 'entropy': '0.02876', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '12.75', 'epoch': '0.336'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:40:25] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             84%|████████▍ | 42/50 [09:57<01:54, 14.35s/it]#015 86%|████████▌ |                    \n",
+       "                             43/50 [10:09<01:34, 13.51s/it]#015                                                    \n",
+       "                             #015{'loss': '0.003902', 'grad_norm': '3.486', 'learning_rate':                       \n",
+       "                             '1.6e-09', 'num_tokens': '2.784e+06', 'completions/mean_length':                      \n",
+       "                             '26.62', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '42', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '26.62',                                        \n",
+       "                             'completions/min_terminated_length': '20',                                            \n",
+       "                             'completions/max_terminated_length': '42',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7734',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4203',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9293',                                               \n",
+       "                             'rewards/format_reward/std': '0.258', 'reward': '1.238',                              \n",
+       "                             'reward_std': '0.4983', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.0002812', 'entropy': '0.01698', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '11.52', 'epoch': '0.344'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:40:25]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027843;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027844;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m84\u001b[0m%|████████▍ | \u001b[1m42\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:57<01:54, \u001b[1m14.\u001b[0m35s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m86\u001b[0m%|████████▌ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m43\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:09<01:34, \u001b[1m13.\u001b[0m51s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.003902', 'grad_norm': '3.486', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '1.6e-09', 'num_tokens': '2.784e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '26.62', 'completions/min_length': '20', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '42', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '26.62', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '20', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '42', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.7734', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4203', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9293', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.258', 'reward': '1.238', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.4983', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.0002812', 'entropy': '0.01698', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '11.52', 'epoch': '0.344'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:40:41] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             86%|████████▌ | 43/50 [10:09<01:34, 13.51s/it]#015 88%|████████▊ |                    \n",
+       "                             44/50 [10:24<01:23, 13.88s/it]#015                                                    \n",
+       "                             #015{'loss': '-0.001483', 'grad_norm': '1.746', 'learning_rate':                      \n",
+       "                             '1.4e-09', 'num_tokens': '2.851e+06', 'completions/mean_length':                      \n",
+       "                             '28.79', 'completions/min_length': '21', 'completions/max_length':                    \n",
+       "                             '40', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '28.79',                                        \n",
+       "                             'completions/min_terminated_length': '21',                                            \n",
+       "                             'completions/max_terminated_length': '40',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7266',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4475',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9451',                                               \n",
+       "                             'rewards/format_reward/std': '0.2294', 'reward': '1.199',                             \n",
+       "                             'reward_std': '0.5036', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.0003025', 'entropy': '0.03641', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '14.75', 'epoch': '0.352'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:40:41]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027851;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027852;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m86\u001b[0m%|████████▌ | \u001b[1m43\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:09<01:34, \u001b[1m13.\u001b[0m51s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m88\u001b[0m%|████████▊ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m44\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:24<01:23, \u001b[1m13.\u001b[0m88s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.001483', 'grad_norm': '1.746', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '1.4e-09', 'num_tokens': '2.851e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '28.79', 'completions/min_length': '21', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '40', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '28.79', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '21', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '40', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.7266', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4475', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9451', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.2294', 'reward': '1.199', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.5036', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.0003025', 'entropy': '0.03641', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '14.75', 'epoch': '0.352'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:40:51] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             88%|████████▊ | 44/50 [10:24<01:23, 13.88s/it]#015 90%|█████████ |                    \n",
+       "                             45/50 [10:35<01:06, 13.20s/it]#015                                                    \n",
+       "                             #015{'loss': '-0.005117', 'grad_norm': '3.015', 'learning_rate':                      \n",
+       "                             '1.2e-09', 'num_tokens': '2.911e+06', 'completions/mean_length':                      \n",
+       "                             '32.46', 'completions/min_length': '21', 'completions/max_length':                    \n",
+       "                             '44', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '32.46',                                        \n",
+       "                             'completions/min_terminated_length': '21',                                            \n",
+       "                             'completions/max_terminated_length': '44',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.8828',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.3229',                                             \n",
+       "                             'rewards/format_reward/mean': '0.8979',                                               \n",
+       "                             'rewards/format_reward/std': '0.3049', 'reward': '1.332',                             \n",
+       "                             'reward_std': '0.4673', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '4.154e-05', 'entropy': '0.009141', 'clip_ratio/low_mean': '0',                       \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '11.61', 'epoch': '0.36'}                                                \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:40:51]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027859;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027860;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m88\u001b[0m%|████████▊ | \u001b[1m44\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:24<01:23, \u001b[1m13.\u001b[0m88s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m90\u001b[0m%|█████████ | \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m45\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:35<01:06, \u001b[1m13.\u001b[0m20s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.005117', 'grad_norm': '3.015', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '1.2e-09', 'num_tokens': '2.911e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '32.46', 'completions/min_length': '21', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '44', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '32.46', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '21', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '44', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.8828', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.3229', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.8979', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.3049', 'reward': '1.332', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.4673', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '4.154e-05', 'entropy': '0.009141', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '11.61', 'epoch': '0.36'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:41:07] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             90%|█████████ | 45/50 [10:35<01:06, 13.20s/it]#015 92%|█████████▏|                    \n",
+       "                             46/50 [10:48<00:52, 13.04s/it]#015                                                    \n",
+       "                             #015{'loss': '0.01022', 'grad_norm': '3.074', 'learning_rate':                        \n",
+       "                             '1e-09', 'num_tokens': '2.974e+06', 'completions/mean_length':                        \n",
+       "                             '34.24', 'completions/min_length': '21', 'completions/max_length':                    \n",
+       "                             '67', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '34.24',                                        \n",
+       "                             'completions/min_terminated_length': '21',                                            \n",
+       "                             'completions/max_terminated_length': '67',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.5703',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.497',                                              \n",
+       "                             'rewards/format_reward/mean': '0.9293',                                               \n",
+       "                             'rewards/format_reward/std': '0.258', 'reward': '1.035',                              \n",
+       "                             'reward_std': '0.5516', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.0002523', 'entropy': '0.0557', 'clip_ratio/low_mean': '0',                         \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '12.63', 'epoch': '0.368'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:41:07]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027867;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027868;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m90\u001b[0m%|█████████ | \u001b[1m45\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:35<01:06, \u001b[1m13.\u001b[0m20s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m92\u001b[0m%|█████████▏| \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m46\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:48<00:52, \u001b[1m13.\u001b[0m04s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.01022', 'grad_norm': '3.074', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '1e-09', 'num_tokens': '2.974e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '34.24', 'completions/min_length': '21', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '67', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '34.24', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '21', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '67', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.5703', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.497', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9293', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.258', 'reward': '1.035', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.5516', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.0002523', 'entropy': '0.0557', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '12.63', 'epoch': '0.368'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:41:18] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             92%|█████████▏| 46/50 [10:48<00:52, 13.04s/it]#015 94%|█████████▍|                    \n",
+       "                             47/50 [11:03<00:40, 13.49s/it]#015                                                    \n",
+       "                             #015{'loss': '-0.005655', 'grad_norm': '2.675', 'learning_rate':                      \n",
+       "                             '8e-10', 'num_tokens': '3.033e+06', 'completions/mean_length':                        \n",
+       "                             '33.34', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '55', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '33.34',                                        \n",
+       "                             'completions/min_terminated_length': '20',                                            \n",
+       "                             'completions/max_terminated_length': '55',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7578',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4301',                                             \n",
+       "                             'rewards/format_reward/mean': '0.8978',                                               \n",
+       "                             'rewards/format_reward/std': '0.3051', 'reward': '1.207',                             \n",
+       "                             'reward_std': '0.5351', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '5.591e-05', 'entropy': '0.01559', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '14.54', 'epoch': '0.376'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:41:18]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027875;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027876;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m92\u001b[0m%|█████████▏| \u001b[1m46\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:48<00:52, \u001b[1m13.\u001b[0m04s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m94\u001b[0m%|█████████▍| \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m47\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:03<00:40, \u001b[1m13.\u001b[0m49s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.005655', 'grad_norm': '2.675', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '8e-10', 'num_tokens': '3.033e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '33.34', 'completions/min_length': '20', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '55', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '33.34', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '20', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '55', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.7578', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4301', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.8978', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.3051', 'reward': '1.207', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.5351', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '5.591e-05', 'entropy': '0.01559', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '14.54', 'epoch': '0.376'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:41:34] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             94%|█████████▍| 47/50 [11:03<00:40, 13.49s/it]#015 96%|█████████▌|                    \n",
+       "                             48/50 [11:19<00:28, 14.30s/it]#015                                                    \n",
+       "                             #015{'loss': '0.004514', 'grad_norm': '1.322', 'learning_rate':                       \n",
+       "                             '6e-10', 'num_tokens': '3.098e+06', 'completions/mean_length':                        \n",
+       "                             '37.12', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '128', 'completions/clipped_ratio': '0.05469',                                        \n",
+       "                             'completions/mean_terminated_length': '31.86',                                        \n",
+       "                             'completions/min_terminated_length': '20',                                            \n",
+       "                             'completions/max_terminated_length': '102',                                           \n",
+       "                             'rewards/tool_call_reward/mean': '0.7656',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4253',                                             \n",
+       "                             'rewards/format_reward/mean': '0.8895',                                               \n",
+       "                             'rewards/format_reward/std': '0.3165', 'reward': '1.21',                              \n",
+       "                             'reward_std': '0.5396', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.0001637', 'entropy': '0.01912', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '16.17', 'epoch': '0.384'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:41:34]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027883;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027884;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m94\u001b[0m%|█████████▍| \u001b[1m47\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:03<00:40, \u001b[1m13.\u001b[0m49s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m96\u001b[0m%|█████████▌| \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m48\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:19<00:28, \u001b[1m14.\u001b[0m30s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.004514', 'grad_norm': '1.322', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '6e-10', 'num_tokens': '3.098e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '37.12', 'completions/min_length': '20', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '128', 'completions/clipped_ratio': '0.05469', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '31.86', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '20', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '102', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.7656', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4253', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.8895', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.3165', 'reward': '1.21', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.5396', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '0.0001637', 'entropy': '0.01912', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '16.17', 'epoch': '0.384'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:41:50] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             96%|█████████▌| 48/50 [11:19<00:28, 14.30s/it]#015 98%|█████████▊|                    \n",
+       "                             49/50 [11:34<00:14, 14.48s/it]#015                                                    \n",
+       "                             #015{'loss': '-0.009328', 'grad_norm': '1.974', 'learning_rate':                      \n",
+       "                             '4e-10', 'num_tokens': '3.167e+06', 'completions/mean_length':                        \n",
+       "                             '36.41', 'completions/min_length': '21', 'completions/max_length':                    \n",
+       "                             '109', 'completions/clipped_ratio': '0',                                              \n",
+       "                             'completions/mean_terminated_length': '36.41',                                        \n",
+       "                             'completions/min_terminated_length': '21',                                            \n",
+       "                             'completions/max_terminated_length': '109',                                           \n",
+       "                             'rewards/tool_call_reward/mean': '0.6953',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4621',                                             \n",
+       "                             'rewards/format_reward/mean': '0.8583',                                               \n",
+       "                             'rewards/format_reward/std': '0.3518', 'reward': '1.124',                             \n",
+       "                             'reward_std': '0.5863', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '3.141e-05', 'entropy': '0.01278', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '14.88', 'epoch': '0.392'}                                               \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:41:50]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027891;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027892;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m96\u001b[0m%|█████████▌| \u001b[1m48\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:19<00:28, \u001b[1m14.\u001b[0m30s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m98\u001b[0m%|█████████▊| \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m49\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:34<00:14, \u001b[1m14.\u001b[0m48s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.009328', 'grad_norm': '1.974', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '4e-10', 'num_tokens': '3.167e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '36.41', 'completions/min_length': '21', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '109', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '36.41', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '21', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '109', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.6953', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.4621', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.8583', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.3518', 'reward': '1.124', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.5863', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '3.141e-05', 'entropy': '0.01278', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '14.88', 'epoch': '0.392'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:42:07] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             98%|█████████▊| 49/50 [11:34<00:14, 14.48s/it]#015100%|██████████|                    \n",
+       "                             50/50 [11:48<00:00, 14.52s/it]#015                                                    \n",
+       "                             #015{'loss': '-0.00214', 'grad_norm': '2.388', 'learning_rate':                       \n",
+       "                             '2e-10', 'num_tokens': '3.236e+06', 'completions/mean_length':                        \n",
+       "                             '29.54', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '46', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '29.54',                                        \n",
+       "                             'completions/min_terminated_length': '20',                                            \n",
+       "                             'completions/max_terminated_length': '46',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.8984',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.3033',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9293',                                               \n",
+       "                             'rewards/format_reward/std': '0.258', 'reward': '1.363',                              \n",
+       "                             'reward_std': '0.4155', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '8.118e-05', 'entropy': '0.01057', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '14.61', 'epoch': '0.4'}                                                 \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:42:07]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027899;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027900;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m98\u001b[0m%|█████████▊| \u001b[1m49\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:34<00:14, \u001b[1m14.\u001b[0m48s/it\u001b[1m]\u001b[0m#\u001b[1m015100\u001b[0m%|██████████| \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m50\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:48<00:00, \u001b[1m14.\u001b[0m52s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.00214', 'grad_norm': '2.388', 'learning_rate': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '2e-10', 'num_tokens': '3.236e+06', 'completions/mean_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '29.54', 'completions/min_length': '20', 'completions/max_length': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '46', 'completions/clipped_ratio': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/mean_terminated_length': '29.54', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/min_terminated_length': '20', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'completions/max_terminated_length': '46', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/mean': '0.8984', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/tool_call_reward/std': '0.3033', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/mean': '0.9293', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'rewards/format_reward/std': '0.258', 'reward': '1.363', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'reward_std': '0.4155', 'frac_reward_zero_std': '0', 'kl': \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m '8.118e-05', 'entropy': '0.01057', 'clip_ratio/low_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'step_time': '14.61', 'epoch': '0.4'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             100%|██████████| 50/50 [11:48<00:00, 14.52s/it]#015                                   \n",
+       "                             #015{'train_runtime': '708.8', 'train_samples_per_second': '9.03',                    \n",
+       "                             'train_steps_per_second': '0.071', 'train_loss': '0.003869',                          \n",
+       "                             'epoch': '0.4'}                                                                       \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027907;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027908;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m100\u001b[0m%|██████████| \u001b[1m50\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:48<00:00, \u001b[1m14.\u001b[0m52s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'train_runtime': '708.8', 'train_samples_per_second': '9.03', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'train_steps_per_second': '0.071', 'train_loss': '0.003869', \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m 'epoch': '0.4'\u001b[1m}\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             100%|██████████| 50/50 [11:48<00:00, 14.52s/it]#015100%|██████████|                   \n",
+       "                             50/50 [11:48<00:00, 14.18s/it]                                                        \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027915;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027916;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m100\u001b[0m%|██████████| \u001b[1m50\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:48<00:00, \u001b[1m14.\u001b[0m52s/it\u001b[1m]\u001b[0m#\u001b[1m015100\u001b[0m%|██████████| \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m \u001b[1m50\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:48<00:00, \u001b[1m14.\u001b[0m18s/it\u001b[1m]\u001b[0m \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:42:12] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             Writing model shards:   0%|          | 0/1 [00:00<?,                                  \n",
+       "                             ?it/s]#015Writing model shards: 100%|██████████| 1/1 [00:02<00:00,                    \n",
+       "                             2.70s/it]#015Writing model shards: 100%|██████████| 1/1                               \n",
+       "                             [00:02<00:00,  2.70s/it]                                                              \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:42:12]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027923;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027924;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Writing model shards: \u001b[1m0\u001b[0m%| | \u001b[1m0\u001b[0m/\u001b[1m1\u001b[0m \u001b[1m[\u001b[0m00:00 INFO smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n", + " 09: \n", + " ++ echo 'Training Container Execution Completed' \n", + "\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027931;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027932;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m ++ echo 'Training Container Execution Completed' \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
+       "                             09:                                                                                   \n",
+       "                             Training Container Execution Completed                                                \n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027939;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027940;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n", + "\u001b[2m \u001b[0m \u001b[1m09\u001b[0m: \u001b[2m \u001b[0m\n", + "\u001b[2m \u001b[0m Training Container Execution Completed \u001b[2m \u001b[0m\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
[06/26/26 10:42:44] INFO     Final Resource Status: Completed                                    resources.py:31442\n",
+       "
\n" + ], + "text/plain": [ + "\u001b[2m[06/26/26 10:42:44]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO \u001b[0m Final Resource Status: \u001b[1mCompleted\u001b[0m \u001b]8;id=17027947;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027948;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31442\u001b\\\u001b[2m31442\u001b[0m\u001b]8;;\u001b\\\n" + ] + }, + "metadata": {}, + "output_type": "display_data" + }, + { + "data": { + "text/html": [ + "
\n"
+      ],
+      "text/plain": []
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    }
+   ],
    "source": [
     "import time\n",
     "from sagemaker.train.model_trainer import ModelTrainer\n",
@@ -717,11 +6563,11 @@
     "    hyperparameters={\n",
     "        \"model_id\": MODEL_ID,\n",
     "        \"max_steps\": 20 if DEBUG_NO_UPDATE else 50,\n",
-    "        \"per_device_train_batch_size\": 1,\n",
-    "        \"gradient_accumulation_steps\": 8,      # global batch = 4 GPUs x 1 x 8\n",
+    "        \"per_device_train_batch_size\": 2,\n",
+    "        \"gradient_accumulation_steps\": 8,      # global batch = 8 GPUs x 2 x 8\n",
     "        \"num_generations\": 8,\n",
     "        \"learning_rate\": 0.0 if DEBUG_NO_UPDATE else 1e-8,\n",
-    "        \"beta\": 0.0 if DEBUG_NO_UPDATE else 0.1,\n",
+    "        \"beta\": 0.1,                            # keep the reference model + KL term enabled\n",
     "        \"temperature\": 0.7,\n",
     "        \"top_p\": 1.0,\n",
     "        \"top_k\": 0,\n",
@@ -729,13 +6575,13 @@
     "        \"max_completion_length\": 128,\n",
     "        \"reward_weights\": \"1.0,0.5\",          # exact match + lighter format shaping\n",
     "        \"repetition_penalty\": 1.05,            # discourage max-length loops\n",
-    "        \"log_completions\": True,\n",
+    "        \"log_completions\": False,\n",
     "        \"num_completions_to_print\": 8,\n",
     "        \"stop_on_collapse\": True,\n",
     "        \"remove_invalid_values\": False,\n",
     "        \"renormalize_logits\": True,\n",
     "        \"gradient_checkpointing\": True,\n",
-    "        \"deepspeed\": \"ds_zero2.json\",          # shard optimizer state across the 4 GPUs (required to fit)\n",
+    "        \"deepspeed\": \"ds_zero3.json\",          # shard model + optimizer state across the 8 GPUs\n",
     "    },\n",
     ")\n",
     "\n",
@@ -957,6 +6803,7 @@
   },
   {
    "cell_type": "markdown",
+   "id": "5338ef32",
    "metadata": {},
    "source": [
     "## Conclusion\n",

From 93b429fb15b5cc268f8d996ab7c3dd6df59b8967 Mon Sep 17 00:00:00 2001
From: DWarez 
Date: Fri, 26 Jun 2026 11:18:09 +0200
Subject: [PATCH 3/5] change: revert back to default Sagemaker launcher +
 sagemaker torchrun

The workaround I did before is totally unnecessary for ml.p4d.24xlarge
therefore I think it was due to some hardware related issue (like high
memory pressure or comms bandwidth). Since we're keeping ml.p4d.24xlarge
as instance type in order to also run the reference model, we can safely
revert back to not use the workaround

Signed-off-by: DWarez 
---
 .../grpo-llm-trl/sagemaker-notebook.ipynb     | 4293 ++++++++---------
 1 file changed, 2142 insertions(+), 2151 deletions(-)

diff --git a/docs/sagemaker/notebooks/sagemaker-sdk/grpo-llm-trl/sagemaker-notebook.ipynb b/docs/sagemaker/notebooks/sagemaker-sdk/grpo-llm-trl/sagemaker-notebook.ipynb
index d6cf4c6da..51353e2e8 100644
--- a/docs/sagemaker/notebooks/sagemaker-sdk/grpo-llm-trl/sagemaker-notebook.ipynb
+++ b/docs/sagemaker/notebooks/sagemaker-sdk/grpo-llm-trl/sagemaker-notebook.ipynb
@@ -454,7 +454,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 24,
+   "execution_count": 28,
    "id": "f42a218b",
    "metadata": {},
    "outputs": [
@@ -635,30 +635,6 @@
     "ZeRO-3 shards the model, gradients, and optimizer state across the eight A100 GPUs. This run keeps `beta` enabled, so TRL also loads a reference model for the KL term.\n"
    ]
   },
-  {
-   "cell_type": "code",
-   "execution_count": 25,
-   "id": "16d3f677",
-   "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Overwriting src/launch.sh\n"
-     ]
-    }
-   ],
-   "source": [
-    "%%writefile src/launch.sh\n",
-    "#!/bin/sh\n",
-    "set -eu\n",
-    "\n",
-    "# Start one training process per GPU; SageMaker exposes the count as SM_NUM_GPUS.\n",
-    "# unset NCCL_PROTO\n",
-    "exec torchrun --nnodes=1 --nproc_per_node=\"${SM_NUM_GPUS:-4}\" train.py \"$@\"\n"
-   ]
-  },
   {
    "cell_type": "code",
    "execution_count": null,
@@ -692,13 +668,7 @@
    "source": [
     "## Launch the training job\n",
     "\n",
-    "Multi-GPU GRPO needs one process per GPU. `launch.sh` is the SageMaker entry point because it can start `torchrun` with the GPU count SageMaker exposes:\n",
-    "\n",
-    "```sh\n",
-    "torchrun --nnodes=1 --nproc_per_node=\"$SM_NUM_GPUS\" train.py ...\n",
-    "```\n",
-    "\n",
-    "`launch.sh` only starts distributed execution; `train.py` contains the training logic.\n",
+    "Multi-GPU GRPO needs one process per GPU. `Torchrun` lets the SageMaker SDK start `train.py` directly on all eight A100 GPUs, so the notebook does not need a shell launcher.\n",
     "\n",
     "This configuration is intentionally small: it is meant to verify that the dataset, reward functions, distributed launch, and logging are wired correctly. Do not expect this short run to materially improve the model; for a real RL training run, increase the number of steps and validate on a held-out set.\n",
     "\n",
@@ -707,19 +677,19 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 26,
+   "execution_count": 29,
    "id": "3204defb",
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:18:12] INFO     OutputDataConfig compression type not provided. Using default:         defaults.py:165\n",
+       "
[06/26/26 10:53:16] INFO     OutputDataConfig compression type not provided. Using default:         defaults.py:165\n",
        "                             GZIP                                                                                  \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:18:12]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m OutputDataConfig compression type not provided. Using default:         \u001b]8;id=17026267;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/train/defaults.py\u001b\\\u001b[2mdefaults.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026268;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/train/defaults.py#165\u001b\\\u001b[2m165\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m[06/26/26 10:53:16]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m OutputDataConfig compression type not provided. Using default:         \u001b]8;id=17027955;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/train/defaults.py\u001b\\\u001b[2mdefaults.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027956;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/train/defaults.py#165\u001b\\\u001b[2m165\u001b[0m\u001b]8;;\u001b\\\n",
        "\u001b[2m                    \u001b[0m         GZIP                                                                   \u001b[2m               \u001b[0m\n"
       ]
      },
@@ -735,7 +705,7 @@
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m Training image URI:                                               \u001b]8;id=17026275;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/train/model_trainer.py\u001b\\\u001b[2mmodel_trainer.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026276;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/train/model_trainer.py#558\u001b\\\u001b[2m558\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m Training image URI:                                               \u001b]8;id=17027963;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/train/model_trainer.py\u001b\\\u001b[2mmodel_trainer.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027964;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/train/model_trainer.py#558\u001b\\\u001b[2m558\u001b[0m\u001b]8;;\u001b\\\n",
        "\u001b[2m                    \u001b[0m         \u001b[1m754289655784.\u001b[0mdkr.ecr.us-east-\u001b[1m1.\u001b[0mamazonaws.com/hf-trl-grpo:sagemake \u001b[2m                    \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         r-trl-dev-e63f67e                                                 \u001b[2m                    \u001b[0m\n"
       ]
@@ -757,7 +727,7 @@
        "\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m SageMaker Python SDK will collect telemetry to help us better \u001b]8;id=17026283;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/telemetry/telemetry_logging.py\u001b\\\u001b[2mtelemetry_logging.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026284;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/telemetry/telemetry_logging.py#110\u001b\\\u001b[2m110\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m SageMaker Python SDK will collect telemetry to help us better \u001b]8;id=17027971;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/telemetry/telemetry_logging.py\u001b\\\u001b[2mtelemetry_logging.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027972;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/telemetry/telemetry_logging.py#110\u001b\\\u001b[2m110\u001b[0m\u001b]8;;\u001b\\\n",
        "\u001b[2m                    \u001b[0m         understand our user's needs, diagnose issues, and deliver     \u001b[2m                        \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         additional features.                                          \u001b[2m                        \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         To opt out of telemetry, please disable via TelemetryOptOut   \u001b[2m                        \u001b[0m\n",
@@ -773,28 +743,11 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:18:19] INFO     Creating training_job resource.                                     resources.py:31116\n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 10:18:19]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m Creating training_job resource.                                     \u001b]8;id=17026291;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026292;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31116\u001b\\\u001b[2m31116\u001b[0m\u001b]8;;\u001b\\\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
/Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/rich/live.py:260: UserWarning: \n",
-       "install \"ipywidgets\" for Jupyter support\n",
-       "  warnings.warn('install \"ipywidgets\" for Jupyter support')\n",
+       "
[06/26/26 10:53:22] INFO     Creating training_job resource.                                     resources.py:31116\n",
        "
\n"
       ],
       "text/plain": [
-       "/Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/rich/live.py:260: UserWarning: \n",
-       "install \"ipywidgets\" for Jupyter support\n",
-       "  warnings.warn('install \"ipywidgets\" for Jupyter support')\n"
+       "\u001b[2m[06/26/26 10:53:22]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m Creating training_job resource.                                     \u001b]8;id=17027979;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027980;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31116\u001b\\\u001b[2m31116\u001b[0m\u001b]8;;\u001b\\\n"
       ]
      },
      "metadata": {},
@@ -803,14 +756,14 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:29:25] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
[06/26/26 11:01:44] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             Starting training script                                                              \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:29:25]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026299;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026300;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:01:44]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17027987;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027988;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         Starting training script                                            \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -820,14 +773,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             ++ /opt/venv/bin/python3 --version                                                    \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026307;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026308;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17027995;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027996;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         ++ /opt/venv/bin/python3 --version                                  \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -837,14 +790,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             Python 3.12.13                                                                        \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026315;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026316;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028003;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028004;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         Python \u001b[1m3.12\u001b[0m.\u001b[1m13\u001b[0m                                                      \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -854,14 +807,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             ++ echo /opt/ml/input/config/resourceconfig.json:                                     \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026323;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026324;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028011;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028012;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         ++ echo /opt/ml/input/config/resourceconfig.json:                   \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -871,15 +824,15 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             ++ cat /opt/ml/input/config/resourceconfig.json                                       \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             /opt/ml/input/config/resourceconfig.json:                                             \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026331;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026332;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         ++ cat /opt/ml/input/config/resourceconfig.json                     \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028019;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028020;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         /opt/ml/input/config/resourceconfig.json:                           \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -888,15 +841,15 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             /opt/ml/input/config/resourceconfig.json:                                             \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             ++ cat /opt/ml/input/config/resourceconfig.json                                       \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026339;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026340;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         /opt/ml/input/config/resourceconfig.json:                           \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028027;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028028;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         ++ cat /opt/ml/input/config/resourceconfig.json                     \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -905,14 +858,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             ++ echo                                                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026347;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026348;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028035;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028036;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         ++ echo                                                             \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -922,8 +875,8 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             {\"current_host\":\"algo-1\",\"current_instance_type\":\"ml.p4d.24xlarge\",                   \n",
        "                             \"current_group_name\":\"homogeneousCluster\",\"hosts\":[\"algo-1\"],\"insta                   \n",
        "                             nce_groups\":[{\"instance_group_name\":\"homogeneousCluster\",\"instance_                   \n",
@@ -932,8 +885,8 @@
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026355;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026356;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028043;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028044;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         \u001b[1m{\u001b[0m\"current_host\":\"algo-1\",\"current_instance_type\":\"ml.p4d.24xlarge\", \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         \"current_group_name\":\"homogeneousCluster\",\"hosts\":\u001b[1m[\u001b[0m\"algo-1\"\u001b[1m]\u001b[0m,\"insta \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         nce_groups\":\u001b[1m[\u001b[0m\u001b[1m{\u001b[0m\"instance_group_name\":\"homogeneousCluster\",\"instance_ \u001b[2m                  \u001b[0m\n",
@@ -947,15 +900,15 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             ++ echo /opt/ml/input/config/inputdataconfig.json:                                    \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             /opt/ml/input/config/inputdataconfig.json:                                            \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026363;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026364;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         ++ echo /opt/ml/input/config/inputdataconfig.json:                  \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028051;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028052;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         /opt/ml/input/config/inputdataconfig.json:                          \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -964,15 +917,15 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             /opt/ml/input/config/inputdataconfig.json:                                            \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             ++ echo /opt/ml/input/config/inputdataconfig.json:                                    \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026371;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026372;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         /opt/ml/input/config/inputdataconfig.json:                          \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028059;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028060;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         ++ echo /opt/ml/input/config/inputdataconfig.json:                  \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -981,14 +934,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             ++ cat /opt/ml/input/config/inputdataconfig.json                                      \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026379;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026380;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028067;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028068;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         ++ cat /opt/ml/input/config/inputdataconfig.json                    \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -998,14 +951,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             ++ echo                                                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026387;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026388;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028075;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028076;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         ++ echo                                                             \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1015,14 +968,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             ++ echo 'Setting up environment variables'                                            \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026395;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026396;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028083;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028084;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         ++ echo 'Setting up environment variables'                          \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1032,23 +985,17 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             {\"code\":{\"TrainingInputMode\":\"File\",\"S3DistributionType\":\"FullyRepl                   \n",
-       "                             icated\",\"RecordWrapperType\":\"None\"},\"sm_drivers\":{\"TrainingInputMod                   \n",
-       "                             e\":\"File\",\"S3DistributionType\":\"FullyReplicated\",\"RecordWrapperType                   \n",
-       "                             \":\"None\"},\"train\":{\"TrainingInputMode\":\"File\",\"S3DistributionType\":                   \n",
-       "                             \"FullyReplicated\",\"RecordWrapperType\":\"None\"}}                                        \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             ++ /opt/venv/bin/python3                                                              \n",
+       "                             /opt/ml/input/data/sm_drivers/scripts/environment.py                                  \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026403;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026404;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m{\u001b[0m\"code\":\u001b[1m{\u001b[0m\"TrainingInputMode\":\"File\",\"S3DistributionType\":\"FullyRepl \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         icated\",\"RecordWrapperType\":\"None\"\u001b[1m}\u001b[0m,\"sm_drivers\":\u001b[1m{\u001b[0m\"TrainingInputMod \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         e\":\"File\",\"S3DistributionType\":\"FullyReplicated\",\"RecordWrapperType \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \":\"None\"\u001b[1m}\u001b[0m,\"train\":\u001b[1m{\u001b[0m\"TrainingInputMode\":\"File\",\"S3DistributionType\": \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"FullyReplicated\",\"RecordWrapperType\":\"None\"\u001b[1m}\u001b[0m\u001b[1m}\u001b[0m                      \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028091;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028092;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         ++ /opt/venv/bin/python3                                            \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         /opt/ml/input/data/sm_drivers/scripts/environment.py                \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -1057,17 +1004,23 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             ++ /opt/venv/bin/python3                                                              \n",
-       "                             /opt/ml/input/data/sm_drivers/scripts/environment.py                                  \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             {\"code\":{\"TrainingInputMode\":\"File\",\"S3DistributionType\":\"FullyRepl                   \n",
+       "                             icated\",\"RecordWrapperType\":\"None\"},\"sm_drivers\":{\"TrainingInputMod                   \n",
+       "                             e\":\"File\",\"S3DistributionType\":\"FullyReplicated\",\"RecordWrapperType                   \n",
+       "                             \":\"None\"},\"train\":{\"TrainingInputMode\":\"File\",\"S3DistributionType\":                   \n",
+       "                             \"FullyReplicated\",\"RecordWrapperType\":\"None\"}}                                        \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026411;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026412;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         ++ /opt/venv/bin/python3                                            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         /opt/ml/input/data/sm_drivers/scripts/environment.py                \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028099;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028100;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m{\u001b[0m\"code\":\u001b[1m{\u001b[0m\"TrainingInputMode\":\"File\",\"S3DistributionType\":\"FullyRepl \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         icated\",\"RecordWrapperType\":\"None\"\u001b[1m}\u001b[0m,\"sm_drivers\":\u001b[1m{\u001b[0m\"TrainingInputMod \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         e\":\"File\",\"S3DistributionType\":\"FullyReplicated\",\"RecordWrapperType \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \":\"None\"\u001b[1m}\u001b[0m,\"train\":\u001b[1m{\u001b[0m\"TrainingInputMode\":\"File\",\"S3DistributionType\": \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \"FullyReplicated\",\"RecordWrapperType\":\"None\"\u001b[1m}\u001b[0m\u001b[1m}\u001b[0m                      \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -1076,14 +1029,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             Setting up environment variables                                                      \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026419;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026420;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028107;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028108;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         Setting up environment variables                                    \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1093,14 +1046,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             No Neurons detected (normal if no neurons installed)                                  \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026427;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026428;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028115;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028116;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         No Neurons detected \u001b[1m(\u001b[0mnormal if no neurons installed\u001b[1m)\u001b[0m                \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1110,14 +1063,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             Environment Variables:                                                                \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026435;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026436;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028123;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028124;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         Environment Variables:                                              \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1127,14 +1080,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             NV_LIBCUBLAS_VERSION=13.1.0.3-1                                                       \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026443;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026444;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028131;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028132;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         NV_LIBCUBLAS_VERSION=13.1.0.3-\u001b[1m1\u001b[0m                                     \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1144,14 +1097,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             NVIDIA_VISIBLE_DEVICES=all                                                            \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026451;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026452;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028139;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028140;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         NVIDIA_VISIBLE_DEVICES=all                                          \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1161,14 +1114,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             PYTHONUNBUFFERED=1                                                                    \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026459;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026460;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028147;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028148;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         PYTHONUNBUFFERED=\u001b[1m1\u001b[0m                                                  \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1178,14 +1131,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             AWS_CONTAINER_CREDENTIALS_RELATIVE_URI=******                                         \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026467;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026468;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028155;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028156;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         AWS_CONTAINER_CREDENTIALS_RELATIVE_URI=******                       \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1195,14 +1148,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SAGEMAKER_TRAINING_MODULE=sagemaker_pytorch_container.training:main                   \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026475;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026476;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028163;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028164;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SAGEMAKER_TRAINING_MODULE=sagemaker_pytorch_container.training:main \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1212,15 +1165,15 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             HOSTNAME=ip-10-0-196-59.ec2.internal                                                  \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             HOSTNAME=ip-10-0-71-109.ec2.internal                                                  \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026483;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026484;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         HOSTNAME=ip-\u001b[1m10\u001b[0m-\u001b[1m0\u001b[0m-\u001b[1m196\u001b[0m-\u001b[1m59.\u001b[0mec2.internal                                \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028171;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028172;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         HOSTNAME=ip-\u001b[1m10\u001b[0m-\u001b[1m0\u001b[0m-\u001b[1m71\u001b[0m-\u001b[1m109.\u001b[0mec2.internal                                \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -1229,8 +1182,8 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             NVIDIA_REQUIRE_CUDA=cuda>=13.0 brand=unknown,driver>=535,driver<536                   \n",
        "                             brand=grid,driver>=535,driver<536                                                     \n",
        "                             brand=tesla,driver>=535,driver<536                                                    \n",
@@ -1284,8 +1237,8 @@
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026491;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026492;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028179;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028180;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         NVIDIA_REQUIRE_CUDA=cuda>=\u001b[1m13.0\u001b[0m brand=unknown,driver>=\u001b[1m535\u001b[0m,driver\u001b[1m<\u001b[0m\u001b[1m536\u001b[0m \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         brand=grid,driver>=\u001b[1m535\u001b[0m,driver<\u001b[1m536\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         brand=tesla,driver>=\u001b[1m535\u001b[0m,driver<\u001b[1m536\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
@@ -1344,14 +1297,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             NV_NVTX_VERSION=13.0.85-1                                                             \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026499;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026500;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028187;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028188;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         NV_NVTX_VERSION=\u001b[1m13\u001b[0m\u001b[1m.0\u001b[0m.\u001b[1m85\u001b[0m-\u001b[1m1\u001b[0m                                           \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1361,14 +1314,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             NV_LIBNPP_VERSION=13.0.1.2-1                                                          \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026507;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026508;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028195;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028196;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         NV_LIBNPP_VERSION=13.0.1.2-\u001b[1m1\u001b[0m                                        \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1378,14 +1331,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             AWS_REGION=us-east-1                                                                  \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026515;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026516;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028203;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028204;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         AWS_REGION=us-east-\u001b[1m1\u001b[0m                                                \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1395,14 +1348,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             PWD=/workspace                                                                        \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026523;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026524;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028211;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028212;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         PWD=/workspace                                                      \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1412,15 +1365,15 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SAGEMAKER_MANAGED_WARMPOOL_CACHE_DIRECTORY=/opt/ml/sagemaker/warmpo                   \n",
        "                             olcache                                                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026531;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026532;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028219;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028220;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SAGEMAKER_MANAGED_WARMPOOL_CACHE_DIRECTORY=/opt/ml/sagemaker/warmpo \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         olcache                                                             \u001b[2m                  \u001b[0m\n"
       ]
@@ -1431,14 +1384,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             NVIDIA_DRIVER_CAPABILITIES=compute,utility,compat32,graphics,video                    \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026539;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026540;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028227;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028228;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         NVIDIA_DRIVER_CAPABILITIES=compute,utility,compat32,graphics,video  \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1448,14 +1401,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             NV_LIBNPP_PACKAGE=libnpp-13-0-13.0.1.2-1                                              \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026547;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026548;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028235;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028236;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         NV_LIBNPP_PACKAGE=libnpp-\u001b[1m13\u001b[0m-\u001b[1m0\u001b[0m-13.0.1.2-\u001b[1m1\u001b[0m                            \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1465,14 +1418,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             NCCL_DEBUG=WARN                                                                       \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026555;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026556;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028243;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028244;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         NCCL_DEBUG=WARN                                                     \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1482,14 +1435,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             NVIDIA_PRODUCT_NAME=CUDA                                                              \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026563;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026564;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028251;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028252;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         NVIDIA_PRODUCT_NAME=CUDA                                            \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1499,14 +1452,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             NV_CUDA_CUDART_VERSION=13.0.96-1                                                      \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026571;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026572;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028259;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028260;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         NV_CUDA_CUDART_VERSION=\u001b[1m13\u001b[0m\u001b[1m.0\u001b[0m.\u001b[1m96\u001b[0m-\u001b[1m1\u001b[0m                                    \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1516,14 +1469,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             HOME=/root                                                                            \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026579;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026580;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028267;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028268;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         HOME=/root                                                          \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1533,14 +1486,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             LANG=C.UTF-8                                                                          \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026587;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026588;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028275;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028276;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         LANG=C.UTF-\u001b[1m8\u001b[0m                                                        \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1550,14 +1503,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             CUDA_VERSION=13.0.2                                                                   \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026595;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026596;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028283;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028284;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         CUDA_VERSION=\u001b[1m13\u001b[0m\u001b[1m.0\u001b[0m.\u001b[1m2\u001b[0m                                                 \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1567,14 +1520,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             HF_HUB_USER_AGENT_ORIGIN=aws:sagemaker:gpu-cuda:training:hf-pytorch                   \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026603;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026604;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028291;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028292;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         HF_HUB_USER_AGENT_ORIGIN=aws:sagemaker:gpu-cuda:training:hf-pytorch \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1584,14 +1537,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             DMLC_INTERFACE=eth0                                                                   \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026611;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026612;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028299;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028300;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         DMLC_INTERFACE=eth0                                                 \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1601,14 +1554,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             HF_TOKEN=******                                                                       \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026619;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026620;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028307;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028308;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         HF_TOKEN=******                                                     \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1618,14 +1571,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             PKG_CMD=yum                                                                           \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026627;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026628;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028315;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028316;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         PKG_CMD=yum                                                         \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1635,14 +1588,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             PYTHONIOENCODING=UTF-8                                                                \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026635;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026636;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028323;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028324;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         PYTHONIOENCODING=UTF-\u001b[1m8\u001b[0m                                              \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1652,14 +1605,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SHLVL=1                                                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026643;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026644;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028331;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028332;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SHLVL=\u001b[1m1\u001b[0m                                                             \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1669,14 +1622,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             NV_CUDA_LIB_VERSION=13.0.2-1                                                          \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026651;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026652;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028339;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028340;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         NV_CUDA_LIB_VERSION=\u001b[1m13\u001b[0m\u001b[1m.0\u001b[0m.\u001b[1m2\u001b[0m-\u001b[1m1\u001b[0m                                        \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1686,14 +1639,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             NVARCH=x86_64                                                                         \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026659;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026660;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028347;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028348;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         NVARCH=x86_64                                                       \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1703,14 +1656,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             PYTHONDONTWRITEBYTECODE=1                                                             \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026667;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026668;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028355;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028356;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         PYTHONDONTWRITEBYTECODE=\u001b[1m1\u001b[0m                                           \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1720,8 +1673,8 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             LD_LIBRARY_PATH=/opt/venv/lib/python3.12/site-packages/nvidia/cudnn                   \n",
        "                             /lib:/opt/venv/lib/python3.12/site-packages/nvidia/nccl/lib:/opt/am                   \n",
        "                             azon/ofi-nccl/lib64:/opt/amazon/openmpi/lib:/opt/amazon/openmpi/lib                   \n",
@@ -1731,8 +1684,8 @@
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026675;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026676;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028363;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028364;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         LD_LIBRARY_PATH=/opt/venv/lib/python3.12/site-packages/nvidia/cudnn \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         /lib:/opt/venv/lib/python3.12/site-packages/nvidia/nccl/lib:/opt/am \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         azon/ofi-nccl/lib64:/opt/amazon/openmpi/lib:/opt/amazon/openmpi/lib \u001b[2m                  \u001b[0m\n",
@@ -1747,17 +1700,17 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             TRAINING_JOB_NAME=smolm3-grpo-toolcall-20260626-081812-202606261018                   \n",
-       "                             12                                                                                    \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             TRAINING_JOB_NAME=smolm3-grpo-toolcall-20260626-085316-202606261053                   \n",
+       "                             16                                                                                    \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026683;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026684;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         TRAINING_JOB_NAME=smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m202606261018\u001b[0m \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m12\u001b[0m                                                                  \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028371;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028372;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         TRAINING_JOB_NAME=smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m202606261053\u001b[0m \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m16\u001b[0m                                                                  \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -1766,14 +1719,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             LC_ALL=C.UTF-8                                                                        \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026691;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026692;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028379;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028380;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         LC_ALL=C.UTF-\u001b[1m8\u001b[0m                                                      \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1783,17 +1736,17 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             TRAINING_JOB_ARN=arn:aws:sagemaker:us-east-1:754289655784:training-                   \n",
-       "                             job/smolm3-grpo-toolcall-20260626-081812-20260626101812                               \n",
+       "                             job/smolm3-grpo-toolcall-20260626-085316-20260626105316                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026699;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026700;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028387;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028388;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         TRAINING_JOB_ARN=arn:aws:sagemaker:us-east-1:754289655784:training- \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         job/smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         job/smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -1802,14 +1755,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             CUDA_HOME=/usr/local/cuda                                                             \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026707;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026708;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028395;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028396;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         CUDA_HOME=/usr/local/cuda                                           \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1819,16 +1772,16 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             PATH=/opt/venv/bin:/opt/amazon/openmpi/bin:/opt/amazon/efa/bin:/usr                   \n",
        "                             /local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/loca                   \n",
        "                             l/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin                                   \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026715;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026716;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028403;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028404;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         PATH=/opt/venv/bin:/opt/amazon/openmpi/bin:/opt/amazon/efa/bin:/usr \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         /local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/loca \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         l/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin                 \u001b[2m                  \u001b[0m\n"
@@ -1840,14 +1793,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             UV_PROJECT_ENVIRONMENT=/opt/venv                                                      \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026723;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026724;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028411;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028412;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         UV_PROJECT_ENVIRONMENT=/opt/venv                                    \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1857,14 +1810,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             DLC_CONTAINER_TYPE=training                                                           \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026731;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026732;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028419;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028420;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         DLC_CONTAINER_TYPE=training                                         \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1874,14 +1827,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             _=/opt/venv/bin/python3                                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026739;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026740;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028427;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028428;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         _=/opt/venv/bin/python3                                             \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1891,14 +1844,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
[06/26/26 11:01:45] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_MODEL_DIR=/opt/ml/model                                                            \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026747;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026748;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:01:45]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028435;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028436;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_MODEL_DIR=/opt/ml/model                                          \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1908,14 +1861,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_INPUT_DIR=/opt/ml/input                                                            \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026755;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026756;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028443;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028444;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_IN\u001b[1mPUT\u001b[0m_DIR=/opt/ml/input                                          \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1925,14 +1878,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_INPUT_DATA_DIR=/opt/ml/input/data                                                  \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026763;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026764;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028451;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028452;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_IN\u001b[1mPUT\u001b[0m_DATA_DIR=/opt/ml/input/data                                \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1942,14 +1895,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_INPUT_CONFIG_DIR=/opt/ml/input/config                                              \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026771;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026772;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028459;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028460;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_IN\u001b[1mPUT\u001b[0m_CONFIG_DIR=/opt/ml/input/config                            \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1959,14 +1912,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_OUTPUT_DIR=/opt/ml/output                                                          \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026779;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026780;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028467;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028468;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_OUT\u001b[1mPUT\u001b[0m_DIR=/opt/ml/output                                        \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1976,14 +1929,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_OUTPUT_FAILURE=/opt/ml/output/failure                                              \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026787;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026788;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028475;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028476;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_OUT\u001b[1mPUT\u001b[0m_FAILURE=/opt/ml/output/failure                            \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -1993,14 +1946,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_OUTPUT_DATA_DIR=/opt/ml/output/data                                                \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026795;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026796;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028483;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028484;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_OUT\u001b[1mPUT\u001b[0m_DATA_DIR=/opt/ml/output/data                              \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2010,14 +1963,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_LOG_LEVEL=20                                                                       \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026803;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026804;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028491;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028492;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_LOG_LEVEL=\u001b[1m20\u001b[0m                                                     \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2027,14 +1980,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_MASTER_ADDR=algo-1                                                                 \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026811;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026812;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028499;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028500;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_MASTER_ADDR=algo-\u001b[1m1\u001b[0m                                               \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2044,14 +1997,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_MASTER_PORT=7777                                                                   \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026819;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026820;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028507;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028508;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_MASTER_PORT=\u001b[1m7777\u001b[0m                                                 \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2061,14 +2014,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_SOURCE_DIR=/opt/ml/input/data/code                                                 \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026827;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026828;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028515;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028516;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_SOURCE_DIR=/opt/ml/input/data/code                               \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2078,15 +2031,51 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             SM_ENTRY_SCRIPT=launch.sh                                                             \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             SM_ENTRY_SCRIPT=train.py                                                              \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028523;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028524;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         SM_ENTRY_SCRIPT=train.py                                            \u001b[2m                  \u001b[0m\n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             SM_DISTRIBUTED_DRIVER_DIR=/opt/ml/input/data/sm_drivers/distributed                   \n",
+       "                             _drivers                                                                              \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028531;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028532;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         SM_DISTRIBUTED_DRIVER_DIR=/opt/ml/input/data/sm_drivers/distributed \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         _drivers                                                            \u001b[2m                  \u001b[0m\n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             SM_DISTRIBUTED_CONFIG={\"process_count_per_node\": 8, \"smp\": null}                      \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026835;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026836;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_ENTRY_SCRIPT=launch.sh                                           \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028539;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028540;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         SM_DISTRIBUTED_CONFIG=\u001b[1m{\u001b[0m\"process_count_per_node\": \u001b[1m8\u001b[0m, \"smp\": null\u001b[1m}\u001b[0m    \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -2095,14 +2084,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_CHANNEL_CODE=/opt/ml/input/data/code                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026843;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026844;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028547;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028548;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_CHANNEL_CODE=/opt/ml/input/data/code                             \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2112,14 +2101,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_CHANNEL_SM_DRIVERS=/opt/ml/input/data/sm_drivers                                   \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026851;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026852;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028555;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028556;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_CHANNEL_SM_DRIVERS=/opt/ml/input/data/sm_drivers                 \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2129,14 +2118,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_CHANNEL_TRAIN=/opt/ml/input/data/train                                             \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026859;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026860;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028563;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028564;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_CHANNEL_TRAIN=/opt/ml/input/data/train                           \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2146,14 +2135,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_CHANNELS=['code', 'sm_drivers', 'train']                                           \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026867;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026868;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028571;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028572;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_CHANNELS=\u001b[1m[\u001b[0m'code', 'sm_drivers', 'train'\u001b[1m]\u001b[0m                         \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2163,14 +2152,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_HP_BETA=0.1                                                                        \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026875;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026876;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028579;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028580;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_HP_BETA=\u001b[1m0\u001b[0m\u001b[1m.1\u001b[0m                                                      \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2180,14 +2169,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_HP_DEEPSPEED=ds_zero3.json                                                         \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026883;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026884;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028587;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028588;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_HP_DEEPSPEED=ds_zero3.json                                       \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2197,14 +2186,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_HP_GRADIENT_ACCUMULATION_STEPS=8                                                   \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026891;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026892;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028595;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028596;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_HP_GRADIENT_ACCUMULATION_STEPS=\u001b[1m8\u001b[0m                                 \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2214,14 +2203,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_HP_GRADIENT_CHECKPOINTING=True                                                     \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026899;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026900;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028603;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028604;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_HP_GRADIENT_CHECKPOINTING=\u001b[3mTrue\u001b[0m                                   \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2231,14 +2220,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_HP_LEARNING_RATE=1e-08                                                             \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026907;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026908;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028611;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028612;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_HP_LEARNING_RATE=\u001b[1m1e\u001b[0m\u001b[1m-08\u001b[0m                                           \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2248,14 +2237,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_HP_LOG_COMPLETIONS=False                                                           \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026915;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026916;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028619;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028620;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_HP_LOG_COMPLETIONS=\u001b[3mFalse\u001b[0m                                         \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2265,14 +2254,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_HP_MAX_COMPLETION_LENGTH=128                                                       \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026923;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026924;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028627;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028628;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_HP_MAX_COMPLETION_LENGTH=\u001b[1m128\u001b[0m                                     \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2282,14 +2271,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_HP_MAX_GRAD_NORM=0.05                                                              \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026931;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026932;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028635;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028636;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_HP_MAX_GRAD_NORM=\u001b[1m0\u001b[0m\u001b[1m.05\u001b[0m                                            \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2299,14 +2288,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_HP_MAX_STEPS=50                                                                    \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026939;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026940;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028643;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028644;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_HP_MAX_STEPS=\u001b[1m50\u001b[0m                                                  \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2316,14 +2305,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_HP_MODEL_ID=HuggingFaceTB/SmolLM3-3B                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026947;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026948;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028651;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028652;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_HP_MODEL_ID=HuggingFaceTB/SmolLM3-3B                             \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2333,14 +2322,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_HP_NUM_COMPLETIONS_TO_PRINT=8                                                      \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026955;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026956;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028659;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028660;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_HP_NUM_COMPLETIONS_TO_PRINT=\u001b[1m8\u001b[0m                                    \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2350,14 +2339,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_HP_NUM_GENERATIONS=8                                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026963;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026964;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028667;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028668;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_HP_NUM_GENERATIONS=\u001b[1m8\u001b[0m                                             \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2367,14 +2356,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_HP_PER_DEVICE_TRAIN_BATCH_SIZE=2                                                   \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026971;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026972;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028675;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028676;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_HP_PER_DEVICE_TRAIN_BATCH_SIZE=\u001b[1m2\u001b[0m                                 \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2384,14 +2373,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_HP_REMOVE_INVALID_VALUES=False                                                     \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026979;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026980;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028683;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028684;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_HP_REMOVE_INVALID_VALUES=\u001b[3mFalse\u001b[0m                                   \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2401,14 +2390,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_HP_RENORMALIZE_LOGITS=True                                                         \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026987;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026988;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028691;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028692;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_HP_RENORMALIZE_LOGITS=\u001b[3mTrue\u001b[0m                                       \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2418,14 +2407,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_HP_REPETITION_PENALTY=1.05                                                         \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17026995;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17026996;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028699;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028700;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_HP_REPETITION_PENALTY=\u001b[1m1\u001b[0m\u001b[1m.05\u001b[0m                                       \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2435,14 +2424,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_HP_REWARD_WEIGHTS=1.0,0.5                                                          \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027003;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027004;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028707;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028708;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_HP_REWARD_WEIGHTS=\u001b[1m1\u001b[0m\u001b[1m.0\u001b[0m,\u001b[1m0.5\u001b[0m                                        \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2452,14 +2441,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_HP_STOP_ON_COLLAPSE=True                                                           \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027011;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027012;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028715;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028716;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_HP_STOP_ON_COLLAPSE=\u001b[3mTrue\u001b[0m                                         \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2469,14 +2458,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_HP_TEMPERATURE=0.7                                                                 \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027019;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027020;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028723;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028724;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_HP_TEMPERATURE=\u001b[1m0\u001b[0m\u001b[1m.7\u001b[0m                                               \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2486,14 +2475,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_HP_TOP_K=0                                                                         \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027027;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027028;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028731;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028732;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_HP_TOP_K=\u001b[1m0\u001b[0m                                                       \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2503,14 +2492,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_HP_TOP_P=1.0                                                                       \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027035;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027036;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028739;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028740;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_HP_TOP_P=\u001b[1m1\u001b[0m\u001b[1m.0\u001b[0m                                                     \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2520,8 +2509,8 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_HPS={\"beta\": 0.1, \"deepspeed\": \"ds_zero3.json\",                                    \n",
        "                             \"gradient_accumulation_steps\": 8, \"gradient_checkpointing\": true,                     \n",
        "                             \"learning_rate\": 1e-08, \"log_completions\": false,                                     \n",
@@ -2535,8 +2524,8 @@
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027043;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027044;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028747;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028748;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_HPS=\u001b[1m{\u001b[0m\"beta\": \u001b[1m0.1\u001b[0m, \"deepspeed\": \"ds_zero3.json\",                  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         \"gradient_accumulation_steps\": \u001b[1m8\u001b[0m, \"gradient_checkpointing\": true,   \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         \"learning_rate\": \u001b[1m1e-08\u001b[0m, \"log_completions\": false,                   \u001b[2m                  \u001b[0m\n",
@@ -2555,14 +2544,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_CURRENT_HOST=algo-1                                                                \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027051;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027052;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028755;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028756;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_CURRENT_HOST=algo-\u001b[1m1\u001b[0m                                              \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2572,14 +2561,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_CURRENT_INSTANCE_TYPE=ml.p4d.24xlarge                                              \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027059;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027060;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028763;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028764;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_CURRENT_INSTANCE_TYPE=ml.p4d.24xlarge                            \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2589,14 +2578,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_HOSTS=['algo-1']                                                                   \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027067;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027068;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028771;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028772;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_HOSTS=\u001b[1m[\u001b[0m'algo-1'\u001b[1m]\u001b[0m                                                 \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2606,14 +2595,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_NETWORK_INTERFACE_NAME=eth0                                                        \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027075;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027076;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028779;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028780;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_NETWORK_INTERFACE_NAME=eth0                                      \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2623,14 +2612,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_HOST_COUNT=1                                                                       \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027083;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027084;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028787;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028788;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_HOST_COUNT=\u001b[1m1\u001b[0m                                                     \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2640,14 +2629,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_CURRENT_HOST_RANK=0                                                                \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027091;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027092;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028795;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028796;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_CURRENT_HOST_RANK=\u001b[1m0\u001b[0m                                              \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2657,14 +2646,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_NUM_CPUS=96                                                                        \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027099;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027100;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028803;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028804;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_NUM_CPUS=\u001b[1m96\u001b[0m                                                      \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2674,14 +2663,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_NUM_GPUS=8                                                                         \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027107;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027108;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028811;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028812;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_NUM_GPUS=\u001b[1m8\u001b[0m                                                       \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2691,14 +2680,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_NUM_NEURONS=0                                                                      \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027115;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027116;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028819;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028820;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_NUM_NEURONS=\u001b[1m0\u001b[0m                                                    \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2708,8 +2697,8 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_RESOURCE_CONFIG={\"current_host\": \"algo-1\",                                         \n",
        "                             \"current_instance_type\": \"ml.p4d.24xlarge\", \"current_group_name\":                     \n",
        "                             \"homogeneousCluster\", \"hosts\": [\"algo-1\"], \"instance_groups\":                         \n",
@@ -2719,8 +2708,8 @@
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027123;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027124;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028827;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028828;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_RESOURCE_CONFIG=\u001b[1m{\u001b[0m\"current_host\": \"algo-1\",                       \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         \"current_instance_type\": \"ml.p4d.24xlarge\", \"current_group_name\":   \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         \"homogeneousCluster\", \"hosts\": \u001b[1m[\u001b[0m\"algo-1\"\u001b[1m]\u001b[0m, \"instance_groups\":       \u001b[2m                  \u001b[0m\n",
@@ -2735,8 +2724,8 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_INPUT_DATA_CONFIG={\"code\": {\"TrainingInputMode\": \"File\",                           \n",
        "                             \"S3DistributionType\": \"FullyReplicated\", \"RecordWrapperType\":                         \n",
        "                             \"None\"}, \"sm_drivers\": {\"TrainingInputMode\": \"File\",                                  \n",
@@ -2747,8 +2736,8 @@
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027131;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027132;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028835;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028836;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_IN\u001b[1mPUT\u001b[0m_DATA_CONFIG=\u001b[1m{\u001b[0m\"code\": \u001b[1m{\u001b[0m\"TrainingInputMode\": \"File\",         \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         \"S3DistributionType\": \"FullyReplicated\", \"RecordWrapperType\":       \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         \"None\"\u001b[1m}\u001b[0m, \"sm_drivers\": \u001b[1m{\u001b[0m\"TrainingInputMode\": \"File\",                \u001b[2m                  \u001b[0m\n",
@@ -2764,8 +2753,8 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             SM_TRAINING_ENV={\"channel_input_dirs\": {\"code\":                                       \n",
        "                             \"/opt/ml/input/data/code\", \"sm_drivers\":                                              \n",
        "                             \"/opt/ml/input/data/sm_drivers\", \"train\":                                             \n",
@@ -2790,7 +2779,7 @@
        "                             \"FullyReplicated\", \"RecordWrapperType\": \"None\"}},                                     \n",
        "                             \"input_config_dir\": \"/opt/ml/input/config\", \"input_data_dir\":                         \n",
        "                             \"/opt/ml/input/data\", \"input_dir\": \"/opt/ml/input\", \"job_name\":                       \n",
-       "                             \"smolm3-grpo-toolcall-20260626-081812-20260626101812\", \"log_level\":                   \n",
+       "                             \"smolm3-grpo-toolcall-20260626-085316-20260626105316\", \"log_level\":                   \n",
        "                             20, \"model_dir\": \"/opt/ml/model\", \"network_interface_name\": \"eth0\",                   \n",
        "                             \"num_cpus\": 96, \"num_gpus\": 8, \"num_neurons\": 0, \"output_data_dir\":                   \n",
        "                             \"/opt/ml/output/data\", \"resource_config\": {\"current_host\":                            \n",
@@ -2802,8 +2791,8 @@
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027139;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027140;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028843;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028844;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         SM_TRAINING_ENV=\u001b[1m{\u001b[0m\"channel_input_dirs\": \u001b[1m{\u001b[0m\"code\":                     \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         \"/opt/ml/input/data/code\", \"sm_drivers\":                            \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         \"/opt/ml/input/data/sm_drivers\", \"train\":                           \u001b[2m                  \u001b[0m\n",
@@ -2828,7 +2817,7 @@
        "\u001b[2m                    \u001b[0m         \"FullyReplicated\", \"RecordWrapperType\": \"None\"\u001b[1m}\u001b[0m\u001b[1m}\u001b[0m,                   \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         \"input_config_dir\": \"/opt/ml/input/config\", \"input_data_dir\":       \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         \"/opt/ml/input/data\", \"input_dir\": \"/opt/ml/input\", \"job_name\":     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"smolm3-grpo-toolcall-20260626-081812-20260626101812\", \"log_level\": \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \"smolm3-grpo-toolcall-20260626-085316-20260626105316\", \"log_level\": \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         \u001b[1m20\u001b[0m, \"model_dir\": \"/opt/ml/model\", \"network_interface_name\": \"eth0\", \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         \"num_cpus\": \u001b[1m96\u001b[0m, \"num_gpus\": \u001b[1m8\u001b[0m, \"num_neurons\": \u001b[1m0\u001b[0m, \"output_data_dir\": \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         \"/opt/ml/output/data\", \"resource_config\": \u001b[1m{\u001b[0m\"current_host\":          \u001b[2m                  \u001b[0m\n",
@@ -2845,14 +2834,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             ++ set +x                                                                             \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027147;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027148;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028851;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028852;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         ++ set +x                                                           \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2862,14 +2851,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             ++ cd /opt/ml/input/data/code                                                         \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027155;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027156;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028859;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028860;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         ++ cd /opt/ml/input/data/code                                       \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -2879,15 +2868,15 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             Running Basic Script driver                                                           \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             Running Torchrun Driver                                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027163;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027164;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Running Basic Script driver                                         \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028867;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028868;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         Running Torchrun Driver                                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -2896,15 +2885,15 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             ++ echo 'Running Basic Script driver'                                                 \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             ++ echo 'Running Torchrun Driver'                                                     \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027171;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027172;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         ++ echo 'Running Basic Script driver'                               \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028875;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028876;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         ++ echo 'Running Torchrun Driver'                                   \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -2913,19 +2902,19 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             ++ /opt/venv/bin/python3                                                              \n",
-       "                             /opt/ml/input/data/sm_drivers/distributed_drivers/basic_script_driv                   \n",
-       "                             er.py                                                                                 \n",
+       "                             /opt/ml/input/data/sm_drivers/distributed_drivers/torchrun_driver.p                   \n",
+       "                             y                                                                                     \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027179;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027180;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028883;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028884;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         ++ /opt/venv/bin/python3                                            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         /opt/ml/input/data/sm_drivers/distributed_drivers/basic_script_driv \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         er.py                                                               \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         /opt/ml/input/data/sm_drivers/distributed_drivers/torchrun_driver.p \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         y                                                                   \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -2934,9 +2923,9 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             Executing command: /bin/sh -c chmod +x launch.sh && ./launch.sh                       \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             Executing command: torchrun --nnodes=1 --nproc_per_node=8 train.py                    \n",
        "                             --beta 0.1 --deepspeed ds_zero3.json --gradient_accumulation_steps                    \n",
        "                             8 --gradient_checkpointing true --learning_rate 1e-08                                 \n",
        "                             --log_completions false --max_completion_length 128 --max_grad_norm                   \n",
@@ -2949,9 +2938,9 @@
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027187;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027188;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Executing command: /bin/sh -c chmod +x launch.sh && ./launch.sh     \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028891;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028892;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         Executing command: torchrun --nnodes=\u001b[1m1\u001b[0m --nproc_per_node=\u001b[1m8\u001b[0m train.py  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         --beta \u001b[1m0.1\u001b[0m --deepspeed ds_zero3.json --gradient_accumulation_steps  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         \u001b[1m8\u001b[0m --gradient_checkpointing true --learning_rate \u001b[1m1e-08\u001b[0m               \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         --log_completions false --max_completion_length \u001b[1m128\u001b[0m --max_grad_norm \u001b[2m                  \u001b[0m\n",
@@ -2969,15 +2958,15 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             W0626 08:29:15.449000 26 torch/distributed/run.py:851]                                \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             W0626 09:01:36.952000 89 torch/distributed/run.py:851]                                \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027195;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027196;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         W0626 08:29:15.\u001b[1m449000\u001b[0m \u001b[1m26\u001b[0m torch/distributed/run.py:\u001b[1m851\u001b[0m\u001b[1m]\u001b[0m              \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028899;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028900;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         W0626 09:01:36.\u001b[1m952000\u001b[0m \u001b[1m89\u001b[0m torch/distributed/run.py:\u001b[1m851\u001b[0m\u001b[1m]\u001b[0m              \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -2986,16 +2975,16 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             W0626 08:29:15.449000 26 torch/distributed/run.py:851]                                \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             W0626 09:01:36.952000 89 torch/distributed/run.py:851]                                \n",
        "                             *****************************************                                             \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027203;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027204;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         W0626 08:29:15.\u001b[1m449000\u001b[0m \u001b[1m26\u001b[0m torch/distributed/run.py:\u001b[1m851\u001b[0m\u001b[1m]\u001b[0m              \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028907;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028908;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         W0626 09:01:36.\u001b[1m952000\u001b[0m \u001b[1m89\u001b[0m torch/distributed/run.py:\u001b[1m851\u001b[0m\u001b[1m]\u001b[0m              \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         *****************************************                           \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -3005,18 +2994,18 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             W0626 08:29:15.449000 26 torch/distributed/run.py:851] Setting                        \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             W0626 09:01:36.952000 89 torch/distributed/run.py:851] Setting                        \n",
        "                             OMP_NUM_THREADS environment variable for each process to be 1 in                      \n",
        "                             default, to avoid your system being overloaded, please further tune                   \n",
        "                             the variable for optimal performance in your application as needed.                   \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027211;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027212;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         W0626 08:29:15.\u001b[1m449000\u001b[0m \u001b[1m26\u001b[0m torch/distributed/run.py:\u001b[1m851\u001b[0m\u001b[1m]\u001b[0m Setting      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028915;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028916;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         W0626 09:01:36.\u001b[1m952000\u001b[0m \u001b[1m89\u001b[0m torch/distributed/run.py:\u001b[1m851\u001b[0m\u001b[1m]\u001b[0m Setting      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         OMP_NUM_THREADS environment variable for each process to be \u001b[1m1\u001b[0m in    \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         default, to avoid your system being overloaded, please further tune \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         the variable for optimal performance in your application as needed. \u001b[2m                  \u001b[0m\n"
@@ -3028,16 +3017,16 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             W0626 08:29:15.449000 26 torch/distributed/run.py:851]                                \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             W0626 09:01:36.952000 89 torch/distributed/run.py:851]                                \n",
        "                             *****************************************                                             \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027219;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027220;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         W0626 08:29:15.\u001b[1m449000\u001b[0m \u001b[1m26\u001b[0m torch/distributed/run.py:\u001b[1m851\u001b[0m\u001b[1m]\u001b[0m              \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028923;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028924;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         W0626 09:01:36.\u001b[1m952000\u001b[0m \u001b[1m89\u001b[0m torch/distributed/run.py:\u001b[1m851\u001b[0m\u001b[1m]\u001b[0m              \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         *****************************************                           \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -3047,14 +3036,14 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:29:57] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
[06/26/26 11:02:16] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             df: /root/.triton/autotune: No such file or directory                                 \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:29:57]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027227;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027228;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:02:16]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028931;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028932;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         df: /root/.triton/autotune: No such file or directory               \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -3064,14 +3053,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             NCCL version 2.28.9+cuda13.0                                                          \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027235;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027236;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028939;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028940;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         NCCL version \u001b[1m2.28\u001b[0m.\u001b[1m9\u001b[0m+cuda13.\u001b[1m0\u001b[0m                                        \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -3081,8 +3070,8 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:30:07] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
[06/26/26 11:02:27] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             Fetching 2 files:   0%|          | 0/2 [00:00<?, ?it/s]#015Fetching                   \n",
        "                             2 files:   0%|          | 0/2 [00:00<?, ?it/s]#015Fetching 2 files:                   \n",
        "                             0%|          | 0/2 [00:00<?, ?it/s]#015Fetching 2 files:   0%|                        \n",
@@ -3091,13 +3080,13 @@
        "                             [00:00<?, ?it/s]#015Fetching 2 files:   0%|          | 0/2                            \n",
        "                             [00:00<?, ?it/s]#015Fetching 2 files:   0%|          | 0/2                            \n",
        "                             [00:00<?, ?it/s]#015Fetching 2 files:  50%|█████     | 1/2                            \n",
-       "                             [00:08<00:08,  8.77s/it]#015Fetching 2 files: 100%|██████████| 2/2                    \n",
-       "                             [00:08<00:00,  4.39s/it]                                                              \n",
+       "                             [00:07<00:07,  7.44s/it]#015Fetching 2 files: 100%|██████████| 2/2                    \n",
+       "                             [00:07<00:00,  3.72s/it]                                                              \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:30:07]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027243;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027244;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:02:27]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028947;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028948;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:   \u001b[1m0\u001b[0m%|          | \u001b[1m0\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:00                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             Fetching 2 files:  50%|█████     | 1/2 [00:08<00:08,                                  \n",
-       "                             8.76s/it]#015Fetching 2 files: 100%|██████████| 2/2 [00:08<00:00,                     \n",
-       "                             4.38s/it]                                                                             \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             Fetching 2 files:  50%|█████     | 1/2 [00:07<00:07,                                  \n",
+       "                             7.45s/it]#015Fetching 2 files: 100%|██████████| 2/2 [00:07<00:00,                     \n",
+       "                             3.72s/it]                                                                             \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027251;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027252;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:  \u001b[1m50\u001b[0m%|█████     | \u001b[1m1\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:08<00:08,                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m8.\u001b[0m76s/it\u001b[1m]\u001b[0m#015Fetching \u001b[1m2\u001b[0m files: \u001b[1m100\u001b[0m%|██████████| \u001b[1m2\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:08<00:00,   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m4.\u001b[0m38s/it\u001b[1m]\u001b[0m                                                           \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028955;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028956;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:  \u001b[1m50\u001b[0m%|█████     | \u001b[1m1\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:07<00:07,                \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m7.\u001b[0m45s/it\u001b[1m]\u001b[0m#015Fetching \u001b[1m2\u001b[0m files: \u001b[1m100\u001b[0m%|██████████| \u001b[1m2\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:07<00:00,   \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m3.\u001b[0m72s/it\u001b[1m]\u001b[0m                                                           \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -3137,19 +3126,19 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             Fetching 2 files:  50%|█████     | 1/2 [00:08<00:08,                                  \n",
-       "                             8.76s/it]#015Fetching 2 files: 100%|██████████| 2/2 [00:08<00:00,                     \n",
-       "                             4.38s/it]                                                                             \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             Fetching 2 files:  50%|█████     | 1/2 [00:07<00:07,                                  \n",
+       "                             7.45s/it]#015Fetching 2 files: 100%|██████████| 2/2 [00:07<00:00,                     \n",
+       "                             3.72s/it]                                                                             \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027259;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027260;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:  \u001b[1m50\u001b[0m%|█████     | \u001b[1m1\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:08<00:08,                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m8.\u001b[0m76s/it\u001b[1m]\u001b[0m#015Fetching \u001b[1m2\u001b[0m files: \u001b[1m100\u001b[0m%|██████████| \u001b[1m2\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:08<00:00,   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m4.\u001b[0m38s/it\u001b[1m]\u001b[0m                                                           \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028963;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028964;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:  \u001b[1m50\u001b[0m%|█████     | \u001b[1m1\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:07<00:07,                \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m7.\u001b[0m45s/it\u001b[1m]\u001b[0m#015Fetching \u001b[1m2\u001b[0m files: \u001b[1m100\u001b[0m%|██████████| \u001b[1m2\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:07<00:00,   \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m3.\u001b[0m72s/it\u001b[1m]\u001b[0m                                                           \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -3158,19 +3147,19 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             Fetching 2 files:  50%|█████     | 1/2 [00:08<00:08,                                  \n",
-       "                             8.71s/it]#015Fetching 2 files: 100%|██████████| 2/2 [00:08<00:00,                     \n",
-       "                             4.36s/it]                                                                             \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             Fetching 2 files:  50%|█████     | 1/2 [00:07<00:07,                                  \n",
+       "                             7.45s/it]#015Fetching 2 files: 100%|██████████| 2/2 [00:07<00:00,                     \n",
+       "                             3.73s/it]                                                                             \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027267;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027268;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:  \u001b[1m50\u001b[0m%|█████     | \u001b[1m1\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:08<00:08,                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m8.\u001b[0m71s/it\u001b[1m]\u001b[0m#015Fetching \u001b[1m2\u001b[0m files: \u001b[1m100\u001b[0m%|██████████| \u001b[1m2\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:08<00:00,   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m4.\u001b[0m36s/it\u001b[1m]\u001b[0m                                                           \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028971;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028972;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:  \u001b[1m50\u001b[0m%|█████     | \u001b[1m1\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:07<00:07,                \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m7.\u001b[0m45s/it\u001b[1m]\u001b[0m#015Fetching \u001b[1m2\u001b[0m files: \u001b[1m100\u001b[0m%|██████████| \u001b[1m2\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:07<00:00,   \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m3.\u001b[0m73s/it\u001b[1m]\u001b[0m                                                           \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -3179,19 +3168,19 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             Fetching 2 files:  50%|█████     | 1/2 [00:08<00:08,                                  \n",
-       "                             8.77s/it]#015Fetching 2 files: 100%|██████████| 2/2 [00:08<00:00,                     \n",
-       "                             4.38s/it]                                                                             \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             Fetching 2 files:  50%|█████     | 1/2 [00:07<00:07,                                  \n",
+       "                             7.45s/it]#015Fetching 2 files: 100%|██████████| 2/2 [00:07<00:00,                     \n",
+       "                             3.72s/it]                                                                             \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027275;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027276;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:  \u001b[1m50\u001b[0m%|█████     | \u001b[1m1\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:08<00:08,                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m8.\u001b[0m77s/it\u001b[1m]\u001b[0m#015Fetching \u001b[1m2\u001b[0m files: \u001b[1m100\u001b[0m%|██████████| \u001b[1m2\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:08<00:00,   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m4.\u001b[0m38s/it\u001b[1m]\u001b[0m                                                           \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028979;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028980;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:  \u001b[1m50\u001b[0m%|█████     | \u001b[1m1\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:07<00:07,                \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m7.\u001b[0m45s/it\u001b[1m]\u001b[0m#015Fetching \u001b[1m2\u001b[0m files: \u001b[1m100\u001b[0m%|██████████| \u001b[1m2\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:07<00:00,   \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m3.\u001b[0m72s/it\u001b[1m]\u001b[0m                                                           \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -3200,19 +3189,19 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             Fetching 2 files:  50%|█████     | 1/2 [00:08<00:08,                                  \n",
-       "                             8.77s/it]#015Fetching 2 files: 100%|██████████| 2/2 [00:08<00:00,                     \n",
-       "                             4.38s/it]                                                                             \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             Fetching 2 files:  50%|█████     | 1/2 [00:07<00:07,                                  \n",
+       "                             7.45s/it]#015Fetching 2 files: 100%|██████████| 2/2 [00:07<00:00,                     \n",
+       "                             3.72s/it]                                                                             \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027283;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027284;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:  \u001b[1m50\u001b[0m%|█████     | \u001b[1m1\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:08<00:08,                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m8.\u001b[0m77s/it\u001b[1m]\u001b[0m#015Fetching \u001b[1m2\u001b[0m files: \u001b[1m100\u001b[0m%|██████████| \u001b[1m2\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:08<00:00,   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m4.\u001b[0m38s/it\u001b[1m]\u001b[0m                                                           \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028987;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028988;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:  \u001b[1m50\u001b[0m%|█████     | \u001b[1m1\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:07<00:07,                \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m7.\u001b[0m45s/it\u001b[1m]\u001b[0m#015Fetching \u001b[1m2\u001b[0m files: \u001b[1m100\u001b[0m%|██████████| \u001b[1m2\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:07<00:00,   \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m3.\u001b[0m72s/it\u001b[1m]\u001b[0m                                                           \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -3221,19 +3210,19 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             Fetching 2 files:  50%|█████     | 1/2 [00:08<00:08,                                  \n",
-       "                             8.76s/it]#015Fetching 2 files: 100%|██████████| 2/2 [00:08<00:00,                     \n",
-       "                             4.38s/it]                                                                             \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             Fetching 2 files:  50%|█████     | 1/2 [00:07<00:07,                                  \n",
+       "                             7.46s/it]#015Fetching 2 files: 100%|██████████| 2/2 [00:07<00:00,                     \n",
+       "                             3.73s/it]                                                                             \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027291;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027292;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:  \u001b[1m50\u001b[0m%|█████     | \u001b[1m1\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:08<00:08,                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m8.\u001b[0m76s/it\u001b[1m]\u001b[0m#015Fetching \u001b[1m2\u001b[0m files: \u001b[1m100\u001b[0m%|██████████| \u001b[1m2\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:08<00:00,   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m4.\u001b[0m38s/it\u001b[1m]\u001b[0m                                                           \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028995;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028996;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:  \u001b[1m50\u001b[0m%|█████     | \u001b[1m1\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:07<00:07,                \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m7.\u001b[0m46s/it\u001b[1m]\u001b[0m#015Fetching \u001b[1m2\u001b[0m files: \u001b[1m100\u001b[0m%|██████████| \u001b[1m2\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:07<00:00,   \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m3.\u001b[0m73s/it\u001b[1m]\u001b[0m                                                           \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -3242,19 +3231,19 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             Fetching 2 files:  50%|█████     | 1/2 [00:08<00:08,                                  \n",
-       "                             8.82s/it]#015Fetching 2 files: 100%|██████████| 2/2 [00:08<00:00,                     \n",
-       "                             4.41s/it]                                                                             \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             Fetching 2 files:  50%|█████     | 1/2 [00:07<00:07,                                  \n",
+       "                             7.46s/it]#015Fetching 2 files: 100%|██████████| 2/2 [00:07<00:00,                     \n",
+       "                             3.73s/it]                                                                             \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027299;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027300;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:  \u001b[1m50\u001b[0m%|█████     | \u001b[1m1\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:08<00:08,                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m8.\u001b[0m82s/it\u001b[1m]\u001b[0m#015Fetching \u001b[1m2\u001b[0m files: \u001b[1m100\u001b[0m%|██████████| \u001b[1m2\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:08<00:00,   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m4.\u001b[0m41s/it\u001b[1m]\u001b[0m                                                           \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029003;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029004;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:  \u001b[1m50\u001b[0m%|█████     | \u001b[1m1\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:07<00:07,                \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m7.\u001b[0m46s/it\u001b[1m]\u001b[0m#015Fetching \u001b[1m2\u001b[0m files: \u001b[1m100\u001b[0m%|██████████| \u001b[1m2\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:07<00:00,   \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m3.\u001b[0m73s/it\u001b[1m]\u001b[0m                                                           \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -3263,17 +3252,17 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             Fetching 2 files:   0%|          | 0/2 [00:00<?, ?it/s]#015Fetching                   \n",
-       "                             2 files: 100%|██████████| 2/2 [00:00<00:00, 44150.57it/s]                             \n",
+       "                             2 files: 100%|██████████| 2/2 [00:00<00:00, 39199.10it/s]                             \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027307;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027308;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029011;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029012;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:   \u001b[1m0\u001b[0m%|          | \u001b[1m0\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:00                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             Fetching 2 files:   0%|          | 0/2 [00:00<?, ?it/s]#015Fetching                   \n",
-       "                             2 files: 100%|██████████| 2/2 [00:00<00:00, 41734.37it/s]                             \n",
+       "                             2 files: 100%|██████████| 2/2 [00:00<00:00, 36314.32it/s]                             \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027315;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027316;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029019;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029020;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:   \u001b[1m0\u001b[0m%|          | \u001b[1m0\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:00                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             Fetching 2 files:   0%|          | 0/2 [00:00<?, ?it/s]#015Fetching                   \n",
-       "                             2 files: 100%|██████████| 2/2 [00:00<00:00, 43690.67it/s]                             \n",
+       "                             2 files: 100%|██████████| 2/2 [00:00<00:00, 41120.63it/s]                             \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027323;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027324;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029027;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029028;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:   \u001b[1m0\u001b[0m%|          | \u001b[1m0\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:00                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             Fetching 2 files:   0%|          | 0/2 [00:00<?, ?it/s]#015Fetching                   \n",
-       "                             2 files: 100%|██████████| 2/2 [00:00<00:00, 40329.85it/s]                             \n",
+       "                             2 files: 100%|██████████| 2/2 [00:00<00:00, 41734.37it/s]                             \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027331;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027332;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029035;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029036;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:   \u001b[1m0\u001b[0m%|          | \u001b[1m0\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:00                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             Fetching 2 files:   0%|          | 0/2 [00:00<?, ?it/s]#015Fetching                   \n",
-       "                             2 files: 100%|██████████| 2/2 [00:00<00:00, 37117.73it/s]                             \n",
+       "                             2 files: 100%|██████████| 2/2 [00:00<00:00, 41323.19it/s]                             \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027339;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027340;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029043;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029044;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:   \u001b[1m0\u001b[0m%|          | \u001b[1m0\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:00                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             Fetching 2 files:   0%|          | 0/2 [00:00<?, ?it/s]#015Fetching                   \n",
-       "                             2 files: 100%|██████████| 2/2 [00:00<00:00, 43018.50it/s]                             \n",
+       "                             2 files: 100%|██████████| 2/2 [00:00<00:00, 25497.29it/s]                             \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027347;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027348;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029051;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029052;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:   \u001b[1m0\u001b[0m%|          | \u001b[1m0\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:00                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             Fetching 2 files:   0%|          | 0/2 [00:00<?, ?it/s]#015Fetching                   \n",
-       "                             2 files: 100%|██████████| 2/2 [00:00<00:00, 41120.63it/s]                             \n",
+       "                             2 files: 100%|██████████| 2/2 [00:00<00:00, 39756.44it/s]                             \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027355;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027356;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029059;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029060;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:   \u001b[1m0\u001b[0m%|          | \u001b[1m0\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:00                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             Fetching 2 files:   0%|          | 0/2 [00:00<?, ?it/s]#015Fetching                   \n",
-       "                             2 files: 100%|██████████| 2/2 [00:00<00:00, 39945.75it/s]                             \n",
+       "                             2 files: 100%|██████████| 2/2 [00:00<00:00, 40721.40it/s]                             \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027363;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027364;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029067;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029068;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:   \u001b[1m0\u001b[0m%|          | \u001b[1m0\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:00[06/26/26 10:30:14] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
[06/26/26 11:02:33] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                              The tokenizer has new PAD/BOS/EOS tokens that differ from the                        \n",
        "                             model config and generation config. The model config and generation                   \n",
        "                             config were aligned accordingly, being updated with the tokenizer's                   \n",
@@ -3425,8 +3414,8 @@
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:30:14]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027371;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027372;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:02:33]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029075;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029076;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m          The tokenizer has new PAD/BOS/EOS tokens that differ from the      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         model config and generation config. The model config and generation \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         config were aligned accordingly, being updated with the tokenizer's \u001b[2m                  \u001b[0m\n",
@@ -3440,8 +3429,8 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                              The tokenizer has new PAD/BOS/EOS tokens that differ from the                        \n",
        "                             model config and generation config. The model config and generation                   \n",
        "                             config were aligned accordingly, being updated with the tokenizer's                   \n",
@@ -3450,8 +3439,8 @@
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027379;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027380;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029083;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029084;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m          The tokenizer has new PAD/BOS/EOS tokens that differ from the      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         model config and generation config. The model config and generation \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         config were aligned accordingly, being updated with the tokenizer's \u001b[2m                  \u001b[0m\n",
@@ -3465,8 +3454,8 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                              The tokenizer has new PAD/BOS/EOS tokens that differ from the                        \n",
        "                             model config and generation config. The model config and generation                   \n",
        "                             config were aligned accordingly, being updated with the tokenizer's                   \n",
@@ -3475,8 +3464,8 @@
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027387;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027388;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029091;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029092;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m          The tokenizer has new PAD/BOS/EOS tokens that differ from the      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         model config and generation config. The model config and generation \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         config were aligned accordingly, being updated with the tokenizer's \u001b[2m                  \u001b[0m\n",
@@ -3490,8 +3479,8 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                              The tokenizer has new PAD/BOS/EOS tokens that differ from the                        \n",
        "                             model config and generation config. The model config and generation                   \n",
        "                             config were aligned accordingly, being updated with the tokenizer's                   \n",
@@ -3500,8 +3489,8 @@
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027395;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027396;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029099;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029100;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m          The tokenizer has new PAD/BOS/EOS tokens that differ from the      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         model config and generation config. The model config and generation \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         config were aligned accordingly, being updated with the tokenizer's \u001b[2m                  \u001b[0m\n",
@@ -3515,8 +3504,8 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                              The tokenizer has new PAD/BOS/EOS tokens that differ from the                        \n",
        "                             model config and generation config. The model config and generation                   \n",
        "                             config were aligned accordingly, being updated with the tokenizer's                   \n",
@@ -3525,8 +3514,8 @@
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027403;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027404;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029107;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029108;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m          The tokenizer has new PAD/BOS/EOS tokens that differ from the      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         model config and generation config. The model config and generation \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         config were aligned accordingly, being updated with the tokenizer's \u001b[2m                  \u001b[0m\n",
@@ -3540,8 +3529,8 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                              The tokenizer has new PAD/BOS/EOS tokens that differ from the                        \n",
        "                             model config and generation config. The model config and generation                   \n",
        "                             config were aligned accordingly, being updated with the tokenizer's                   \n",
@@ -3550,8 +3539,8 @@
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027411;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027412;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029115;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029116;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m          The tokenizer has new PAD/BOS/EOS tokens that differ from the      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         model config and generation config. The model config and generation \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         config were aligned accordingly, being updated with the tokenizer's \u001b[2m                  \u001b[0m\n",
@@ -3565,8 +3554,8 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                              The tokenizer has new PAD/BOS/EOS tokens that differ from the                        \n",
        "                             model config and generation config. The model config and generation                   \n",
        "                             config were aligned accordingly, being updated with the tokenizer's                   \n",
@@ -3575,8 +3564,8 @@
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027419;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027420;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029123;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029124;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m          The tokenizer has new PAD/BOS/EOS tokens that differ from the      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         model config and generation config. The model config and generation \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         config were aligned accordingly, being updated with the tokenizer's \u001b[2m                  \u001b[0m\n",
@@ -3590,8 +3579,8 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                              The tokenizer has new PAD/BOS/EOS tokens that differ from the                        \n",
        "                             model config and generation config. The model config and generation                   \n",
        "                             config were aligned accordingly, being updated with the tokenizer's                   \n",
@@ -3600,8 +3589,8 @@
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027427;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027428;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029131;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029132;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m          The tokenizer has new PAD/BOS/EOS tokens that differ from the      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         model config and generation config. The model config and generation \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         config were aligned accordingly, being updated with the tokenizer's \u001b[2m                  \u001b[0m\n",
@@ -3615,16 +3604,16 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             [RANK 0] Gradient accumulation steps mismatch:                                        \n",
        "                             GradientAccumulationPlugin has 1, DeepSpeed config has 8. Using                       \n",
        "                             DeepSpeed's value.                                                                    \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027435;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027436;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029139;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029140;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         \u001b[1m[\u001b[0mRANK \u001b[1m0\u001b[0m\u001b[1m]\u001b[0m Gradient accumulation steps mismatch:                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         GradientAccumulationPlugin has \u001b[1m1\u001b[0m, DeepSpeed config has \u001b[1m8\u001b[0m. Using     \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         DeepSpeed's value.                                                  \u001b[2m                  \u001b[0m\n"
@@ -3636,8 +3625,8 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:30:25] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
[06/26/26 11:02:43] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             0%|          | 0/50 [00:00<?, ?it/s] Ignoring                                         \n",
        "                             clean_up_tokenization_spaces=True for BPE tokenizer                                   \n",
        "                             TokenizersBackend. The clean_up_tokenization post-processing step                     \n",
@@ -3649,8 +3638,8 @@
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:30:25]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027443;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027444;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:02:43]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029147;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029148;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         \u001b[1m0\u001b[0m%|          | \u001b[1m0\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m00:00                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                              Ignoring clean_up_tokenization_spaces=True for BPE tokenizer                         \n",
        "                             TokenizersBackend. The clean_up_tokenization post-processing step                     \n",
        "                             is designed for WordPiece tokenizers and is destructive for BPE (it                   \n",
@@ -3679,8 +3668,8 @@
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027451;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027452;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029155;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029156;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m          Ignoring clean_up_tokenization_spaces=\u001b[3mTrue\u001b[0m for BPE tokenizer       \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         TokenizersBackend. The clean_up_tokenization post-processing step   \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         is designed for WordPiece tokenizers and is destructive for BPE \u001b[1m(\u001b[0mit \u001b[2m                  \u001b[0m\n",
@@ -3696,8 +3685,8 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                              Ignoring clean_up_tokenization_spaces=True for BPE tokenizer                         \n",
        "                             TokenizersBackend. The clean_up_tokenization post-processing step                     \n",
        "                             is designed for WordPiece tokenizers and is destructive for BPE (it                   \n",
@@ -3708,8 +3697,8 @@
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027459;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027460;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029163;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029164;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m          Ignoring clean_up_tokenization_spaces=\u001b[3mTrue\u001b[0m for BPE tokenizer       \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         TokenizersBackend. The clean_up_tokenization post-processing step   \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         is designed for WordPiece tokenizers and is destructive for BPE \u001b[1m(\u001b[0mit \u001b[2m                  \u001b[0m\n",
@@ -3725,8 +3714,8 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                              Ignoring clean_up_tokenization_spaces=True for BPE tokenizer                         \n",
        "                             TokenizersBackend. The clean_up_tokenization post-processing step                     \n",
        "                             is designed for WordPiece tokenizers and is destructive for BPE (it                   \n",
@@ -3737,8 +3726,8 @@
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027467;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027468;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029171;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029172;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m          Ignoring clean_up_tokenization_spaces=\u001b[3mTrue\u001b[0m for BPE tokenizer       \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         TokenizersBackend. The clean_up_tokenization post-processing step   \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         is designed for WordPiece tokenizers and is destructive for BPE \u001b[1m(\u001b[0mit \u001b[2m                  \u001b[0m\n",
@@ -3754,8 +3743,8 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                              Ignoring clean_up_tokenization_spaces=True for BPE tokenizer                         \n",
        "                             TokenizersBackend. The clean_up_tokenization post-processing step                     \n",
        "                             is designed for WordPiece tokenizers and is destructive for BPE (it                   \n",
@@ -3766,8 +3755,8 @@
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027475;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027476;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029179;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029180;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m          Ignoring clean_up_tokenization_spaces=\u001b[3mTrue\u001b[0m for BPE tokenizer       \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         TokenizersBackend. The clean_up_tokenization post-processing step   \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         is designed for WordPiece tokenizers and is destructive for BPE \u001b[1m(\u001b[0mit \u001b[2m                  \u001b[0m\n",
@@ -3783,8 +3772,8 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                              Ignoring clean_up_tokenization_spaces=True for BPE tokenizer                         \n",
        "                             TokenizersBackend. The clean_up_tokenization post-processing step                     \n",
        "                             is designed for WordPiece tokenizers and is destructive for BPE (it                   \n",
@@ -3795,8 +3784,8 @@
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027483;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027484;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029187;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029188;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m          Ignoring clean_up_tokenization_spaces=\u001b[3mTrue\u001b[0m for BPE tokenizer       \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         TokenizersBackend. The clean_up_tokenization post-processing step   \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         is designed for WordPiece tokenizers and is destructive for BPE \u001b[1m(\u001b[0mit \u001b[2m                  \u001b[0m\n",
@@ -3812,8 +3801,8 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                              Ignoring clean_up_tokenization_spaces=True for BPE tokenizer                         \n",
        "                             TokenizersBackend. The clean_up_tokenization post-processing step                     \n",
        "                             is designed for WordPiece tokenizers and is destructive for BPE (it                   \n",
@@ -3824,8 +3813,8 @@
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027491;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027492;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029195;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029196;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m          Ignoring clean_up_tokenization_spaces=\u001b[3mTrue\u001b[0m for BPE tokenizer       \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         TokenizersBackend. The clean_up_tokenization post-processing step   \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         is designed for WordPiece tokenizers and is destructive for BPE \u001b[1m(\u001b[0mit \u001b[2m                  \u001b[0m\n",
@@ -3841,8 +3830,8 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                              Ignoring clean_up_tokenization_spaces=True for BPE tokenizer                         \n",
        "                             TokenizersBackend. The clean_up_tokenization post-processing step                     \n",
        "                             is designed for WordPiece tokenizers and is destructive for BPE (it                   \n",
@@ -3853,8 +3842,8 @@
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027499;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027500;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029203;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029204;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m          Ignoring clean_up_tokenization_spaces=\u001b[3mTrue\u001b[0m for BPE tokenizer       \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         TokenizersBackend. The clean_up_tokenization post-processing step   \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         is designed for WordPiece tokenizers and is destructive for BPE \u001b[1m(\u001b[0mit \u001b[2m                  \u001b[0m\n",
@@ -3870,9 +3859,9 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:30:36] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             2%|▏         | 1/50 [00:19<16:16, 19.92s/it]#015                                      \n",
+       "
[06/26/26 11:02:54] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             2%|▏         | 1/50 [00:20<16:25, 20.10s/it]#015                                      \n",
        "                             #015{'loss': '0.03134', 'grad_norm': '2.996', 'learning_rate':                        \n",
        "                             '1e-08', 'num_tokens': '6.144e+04', 'completions/mean_length':                        \n",
        "                             '30.66', 'completions/min_length': '17', 'completions/max_length':                    \n",
@@ -3887,14 +3876,14 @@
        "                             'frac_reward_zero_std': '0', 'kl': '4.555e-06', 'entropy':                            \n",
        "                             '0.02574', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',                     \n",
        "                             'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',                              \n",
-       "                             'clip_ratio/region_mean': '0', 'step_time': '19.91', 'epoch':                         \n",
+       "                             'clip_ratio/region_mean': '0', 'step_time': '20.09', 'epoch':                         \n",
        "                             '0.008'}                                                                              \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:30:36]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027507;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027508;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m2\u001b[0m%|▏         | \u001b[1m1\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m00:19<16:16, \u001b[1m19.\u001b[0m92s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                    \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:02:54]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029211;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029212;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m2\u001b[0m%|▏         | \u001b[1m1\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m00:20<16:25, \u001b[1m20.\u001b[0m10s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                    \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.03134', 'grad_norm': '2.996', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '1e-08', 'num_tokens': '6.144e+04', 'completions/mean_length':      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '30.66', 'completions/min_length': '17', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
@@ -3909,7 +3898,7 @@
        "\u001b[2m                    \u001b[0m         'frac_reward_zero_std': '0', 'kl': '4.555e-06', 'entropy':          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '0.02574', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',   \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/region_mean': '0', 'step_time': '19.91', 'epoch':       \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'clip_ratio/region_mean': '0', 'step_time': '20.09', 'epoch':       \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '0.008'\u001b[1m}\u001b[0m                                                            \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -3919,15 +3908,15 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:30:52] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             2%|▏         | 1/50 [00:19<16:16, 19.92s/it]#015  4%|▍         |                      \n",
-       "                             2/50 [00:34<13:36, 17.00s/it]#015                                                     \n",
-       "                             #015{'loss': '0.002899', 'grad_norm': '1.937', 'learning_rate':                       \n",
+       "
[06/26/26 11:03:10] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             2%|▏         | 1/50 [00:20<16:25, 20.10s/it]#015  4%|▍         |                      \n",
+       "                             2/50 [00:34<13:37, 17.04s/it]#015                                                     \n",
+       "                             #015{'loss': '0.002893', 'grad_norm': '2.016', 'learning_rate':                       \n",
        "                             '9.8e-09', 'num_tokens': '1.326e+05', 'completions/mean_length':                      \n",
-       "                             '36.66', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '36.78', 'completions/min_length': '20', 'completions/max_length':                    \n",
        "                             '82', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '36.66',                                        \n",
+       "                             'completions/mean_terminated_length': '36.78',                                        \n",
        "                             'completions/min_terminated_length': '20',                                            \n",
        "                             'completions/max_terminated_length': '82',                                            \n",
        "                             'rewards/tool_call_reward/mean': '0.6484',                                            \n",
@@ -3935,22 +3924,22 @@
        "                             'rewards/format_reward/mean': '0.9371',                                               \n",
        "                             'rewards/format_reward/std': '0.2445', 'reward': '1.117',                             \n",
        "                             'reward_std': '0.5346', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '7.544e-05', 'entropy': '0.01716', 'clip_ratio/low_mean': '0',                        \n",
+       "                             '8.694e-05', 'entropy': '0.01816', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '14.94', 'epoch': '0.016'}                                               \n",
+       "                             'step_time': '14.88', 'epoch': '0.016'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:30:52]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027515;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027516;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m2\u001b[0m%|▏         | \u001b[1m1\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m00:19<16:16, \u001b[1m19.\u001b[0m92s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m  \u001b[1m4\u001b[0m%|▍         |    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m2\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m00:34<13:36, \u001b[1m17.\u001b[0m00s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.002899', 'grad_norm': '1.937', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:03:10]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029219;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029220;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m2\u001b[0m%|▏         | \u001b[1m1\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m00:20<16:25, \u001b[1m20.\u001b[0m10s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m  \u001b[1m4\u001b[0m%|▍         |    \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m2\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m00:34<13:37, \u001b[1m17.\u001b[0m04s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.002893', 'grad_norm': '2.016', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '9.8e-09', 'num_tokens': '1.326e+05', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '36.66', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '36.78', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '82', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '36.66',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '36.78',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '82',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.6484',                          \u001b[2m                  \u001b[0m\n",
@@ -3958,10 +3947,10 @@
        "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9371',                             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2445', 'reward': '1.117',           \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'reward_std': '0.5346', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '7.544e-05', 'entropy': '0.01716', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '8.694e-05', 'entropy': '0.01816', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '14.94', 'epoch': '0.016'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '14.88', 'epoch': '0.016'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -3970,49 +3959,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:31:08] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             4%|▍         | 2/50 [00:34<13:36, 17.00s/it]#015  6%|▌         |                      \n",
-       "                             3/50 [00:50<12:57, 16.55s/it]#015                                                     \n",
-       "                             #015{'loss': '-0.0008521', 'grad_norm': '2.254', 'learning_rate':                     \n",
+       "
[06/26/26 11:03:26] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             4%|▍         | 2/50 [00:34<13:37, 17.04s/it]#015  6%|▌         |                      \n",
+       "                             3/50 [00:51<12:59, 16.58s/it]#015                                                     \n",
+       "                             #015{'loss': '-0.0008295', 'grad_norm': '2.204', 'learning_rate':                     \n",
        "                             '9.6e-09', 'num_tokens': '2.039e+05', 'completions/mean_length':                      \n",
        "                             '40.09', 'completions/min_length': '20', 'completions/max_length':                    \n",
        "                             '84', 'completions/clipped_ratio': '0',                                               \n",
        "                             'completions/mean_terminated_length': '40.09',                                        \n",
        "                             'completions/min_terminated_length': '20',                                            \n",
        "                             'completions/max_terminated_length': '84',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7734',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4203',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9371',                                               \n",
-       "                             'rewards/format_reward/std': '0.2445', 'reward': '1.242',                             \n",
-       "                             'reward_std': '0.4905', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.0001991', 'entropy': '0.02427', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'rewards/tool_call_reward/mean': '0.7656',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4253',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9292',                                               \n",
+       "                             'rewards/format_reward/std': '0.2583', 'reward': '1.23',                              \n",
+       "                             'reward_std': '0.5021', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.0002372', 'entropy': '0.02427', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '16', 'epoch': '0.024'}                                                  \n",
+       "                             'step_time': '16.03', 'epoch': '0.024'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:31:08]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027523;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027524;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m4\u001b[0m%|▍         | \u001b[1m2\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m00:34<13:36, \u001b[1m17.\u001b[0m00s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m  \u001b[1m6\u001b[0m%|▌         |    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m3\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m00:50<12:57, \u001b[1m16.\u001b[0m55s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.0008521', 'grad_norm': '2.254', 'learning_rate':   \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:03:26]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029227;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029228;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m4\u001b[0m%|▍         | \u001b[1m2\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m00:34<13:37, \u001b[1m17.\u001b[0m04s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m  \u001b[1m6\u001b[0m%|▌         |    \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m3\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m00:51<12:59, \u001b[1m16.\u001b[0m58s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.0008295', 'grad_norm': '2.204', 'learning_rate':   \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '9.6e-09', 'num_tokens': '2.039e+05', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '40.09', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '84', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '40.09',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '84',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7734',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4203',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9371',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2445', 'reward': '1.242',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.4905', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0001991', 'entropy': '0.02427', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7656',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4253',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9292',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2583', 'reward': '1.23',            \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.5021', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.0002372', 'entropy': '0.02427', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '16', 'epoch': '0.024'\u001b[1m}\u001b[0m                                \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '16.03', 'epoch': '0.024'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -4021,49 +4010,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:31:24] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             6%|▌         | 3/50 [00:50<12:57, 16.55s/it]#015  8%|▊         |                      \n",
-       "                             4/50 [01:04<11:53, 15.50s/it]#015                                                     \n",
-       "                             #015{'loss': '0.0005852', 'grad_norm': '1.884', 'learning_rate':                      \n",
+       "
[06/26/26 11:03:42] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             6%|▌         | 3/50 [00:51<12:59, 16.58s/it]#015  8%|▊         |                      \n",
+       "                             4/50 [01:04<11:53, 15.52s/it]#015                                                     \n",
+       "                             #015{'loss': '0.007161', 'grad_norm': '1.699', 'learning_rate':                       \n",
        "                             '9.4e-09', 'num_tokens': '2.644e+05', 'completions/mean_length':                      \n",
-       "                             '34.04', 'completions/min_length': '22', 'completions/max_length':                    \n",
+       "                             '34.21', 'completions/min_length': '22', 'completions/max_length':                    \n",
        "                             '57', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '34.04',                                        \n",
+       "                             'completions/mean_terminated_length': '34.21',                                        \n",
        "                             'completions/min_terminated_length': '22',                                            \n",
        "                             'completions/max_terminated_length': '57',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.8281',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.3788',                                             \n",
+       "                             'rewards/tool_call_reward/mean': '0.8359',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.3718',                                             \n",
        "                             'rewards/format_reward/mean': '0.9214',                                               \n",
-       "                             'rewards/format_reward/std': '0.2711', 'reward': '1.289',                             \n",
-       "                             'reward_std': '0.4769', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.0001854', 'entropy': '0.01301', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'rewards/format_reward/std': '0.2711', 'reward': '1.297',                             \n",
+       "                             'reward_std': '0.472', 'frac_reward_zero_std': '0', 'kl':                             \n",
+       "                             '6.244e-06', 'entropy': '0.01326', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '13.88', 'epoch': '0.032'}                                               \n",
+       "                             'step_time': '13.87', 'epoch': '0.032'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:31:24]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027531;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027532;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m6\u001b[0m%|▌         | \u001b[1m3\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m00:50<12:57, \u001b[1m16.\u001b[0m55s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m  \u001b[1m8\u001b[0m%|▊         |    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m4\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:04<11:53, \u001b[1m15.\u001b[0m50s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.0005852', 'grad_norm': '1.884', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:03:42]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029235;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029236;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m6\u001b[0m%|▌         | \u001b[1m3\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m00:51<12:59, \u001b[1m16.\u001b[0m58s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m  \u001b[1m8\u001b[0m%|▊         |    \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m4\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:04<11:53, \u001b[1m15.\u001b[0m52s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.007161', 'grad_norm': '1.699', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '9.4e-09', 'num_tokens': '2.644e+05', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '34.04', 'completions/min_length': '22', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '34.21', 'completions/min_length': '22', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '57', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '34.04',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '34.21',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '22',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '57',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8281',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.3788',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8359',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.3718',                           \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9214',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2711', 'reward': '1.289',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.4769', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0001854', 'entropy': '0.01301', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2711', 'reward': '1.297',           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.472', 'frac_reward_zero_std': '0', 'kl':           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '6.244e-06', 'entropy': '0.01326', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '13.88', 'epoch': '0.032'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '13.87', 'epoch': '0.032'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -4072,15 +4061,15 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:31:34] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             8%|▊         | 4/50 [01:04<11:53, 15.50s/it]#015 10%|█         |                      \n",
-       "                             5/50 [01:18<11:01, 14.69s/it]#015                                                     \n",
-       "                             #015{'loss': '-0.004746', 'grad_norm': '2.611', 'learning_rate':                      \n",
+       "
[06/26/26 11:03:53] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             8%|▊         | 4/50 [01:04<11:53, 15.52s/it]#015 10%|█         |                      \n",
+       "                             5/50 [01:17<10:42, 14.28s/it]#015                                                     \n",
+       "                             #015{'loss': '-0.004775', 'grad_norm': '2.619', 'learning_rate':                      \n",
        "                             '9.2e-09', 'num_tokens': '3.378e+05', 'completions/mean_length':                      \n",
-       "                             '35.79', 'completions/min_length': '24', 'completions/max_length':                    \n",
+       "                             '35.8', 'completions/min_length': '24', 'completions/max_length':                     \n",
        "                             '55', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '35.79',                                        \n",
+       "                             'completions/mean_terminated_length': '35.8',                                         \n",
        "                             'completions/min_terminated_length': '24',                                            \n",
        "                             'completions/max_terminated_length': '55',                                            \n",
        "                             'rewards/tool_call_reward/mean': '0.7734',                                            \n",
@@ -4088,22 +4077,22 @@
        "                             'rewards/format_reward/mean': '0.9371',                                               \n",
        "                             'rewards/format_reward/std': '0.2444', 'reward': '1.242',                             \n",
        "                             'reward_std': '0.4905', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '8.713e-05', 'entropy': '0.01761', 'clip_ratio/low_mean': '0',                        \n",
+       "                             '7.967e-05', 'entropy': '0.01727', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '13.24', 'epoch': '0.04'}                                                \n",
+       "                             'step_time': '12.06', 'epoch': '0.04'}                                                \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:31:34]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027539;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027540;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m8\u001b[0m%|▊         | \u001b[1m4\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:04<11:53, \u001b[1m15.\u001b[0m50s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m10\u001b[0m%|█         |    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m5\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:18<11:01, \u001b[1m14.\u001b[0m69s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.004746', 'grad_norm': '2.611', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:03:53]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029243;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029244;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m8\u001b[0m%|▊         | \u001b[1m4\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:04<11:53, \u001b[1m15.\u001b[0m52s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m10\u001b[0m%|█         |    \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m5\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:17<10:42, \u001b[1m14.\u001b[0m28s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.004775', 'grad_norm': '2.619', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '9.2e-09', 'num_tokens': '3.378e+05', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '35.79', 'completions/min_length': '24', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '35.8', 'completions/min_length': '24', 'completions/max_length':   \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '55', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '35.79',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '35.8',                       \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '24',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '55',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7734',                          \u001b[2m                  \u001b[0m\n",
@@ -4111,10 +4100,10 @@
        "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9371',                             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2444', 'reward': '1.242',           \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'reward_std': '0.4905', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '8.713e-05', 'entropy': '0.01761', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '7.967e-05', 'entropy': '0.01727', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '13.24', 'epoch': '0.04'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '12.06', 'epoch': '0.04'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -4123,11 +4112,11 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:31:45] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             10%|█         | 5/50 [01:18<11:01, 14.69s/it]#015 12%|█▏        |                     \n",
-       "                             6/50 [01:30<10:14, 13.97s/it]#015                                                     \n",
-       "                             #015{'loss': '2.203e-07', 'grad_norm': '1.855', 'learning_rate':                      \n",
+       "
[06/26/26 11:04:04] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             10%|█         | 5/50 [01:17<10:42, 14.28s/it]#015 12%|█▏        |                     \n",
+       "                             6/50 [01:29<09:56, 13.55s/it]#015                                                     \n",
+       "                             #015{'loss': '8.66e-07', 'grad_norm': '1.851', 'learning_rate':                       \n",
        "                             '9e-09', 'num_tokens': '4.018e+05', 'completions/mean_length':                        \n",
        "                             '28.62', 'completions/min_length': '20', 'completions/max_length':                    \n",
        "                             '44', 'completions/clipped_ratio': '0',                                               \n",
@@ -4139,18 +4128,18 @@
        "                             'rewards/format_reward/mean': '0.9136',                                               \n",
        "                             'rewards/format_reward/std': '0.2828', 'reward': '1.308',                             \n",
        "                             'reward_std': '0.4707', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '-2.709e-06', 'entropy': '0.01278', 'clip_ratio/low_mean': '0',                       \n",
+       "                             '6.785e-06', 'entropy': '0.01261', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '12.56', 'epoch': '0.048'}                                               \n",
+       "                             'step_time': '12.11', 'epoch': '0.048'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:31:45]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027547;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027548;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m10\u001b[0m%|█         | \u001b[1m5\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:18<11:01, \u001b[1m14.\u001b[0m69s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m12\u001b[0m%|█▏        |   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m6\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:30<10:14, \u001b[1m13.\u001b[0m97s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '2.203e-07', 'grad_norm': '1.855', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:04:04]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029251;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029252;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m10\u001b[0m%|█         | \u001b[1m5\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:17<10:42, \u001b[1m14.\u001b[0m28s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m12\u001b[0m%|█▏        |   \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m6\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:29<09:56, \u001b[1m13.\u001b[0m55s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '8.66e-07', 'grad_norm': '1.851', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '9e-09', 'num_tokens': '4.018e+05', 'completions/mean_length':      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '28.62', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '44', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
@@ -4162,10 +4151,10 @@
        "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9136',                             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2828', 'reward': '1.308',           \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'reward_std': '0.4707', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '-2.709e-06', 'entropy': '0.01278', 'clip_ratio/low_mean': '0',     \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '6.785e-06', 'entropy': '0.01261', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '12.56', 'epoch': '0.048'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '12.11', 'epoch': '0.048'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -4174,49 +4163,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:32:01] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             12%|█▏        | 6/50 [01:30<10:14, 13.97s/it]#015 14%|█▍        |                     \n",
-       "                             7/50 [01:43<09:49, 13.71s/it]#015                                                     \n",
-       "                             #015{'loss': '-0.000197', 'grad_norm': '5.104', 'learning_rate':                      \n",
+       "
[06/26/26 11:04:20] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             12%|█▏        | 6/50 [01:29<09:56, 13.55s/it]#015 14%|█▍        |                     \n",
+       "                             7/50 [01:41<09:24, 13.13s/it]#015                                                     \n",
+       "                             #015{'loss': '-0.0003451', 'grad_norm': '5.135', 'learning_rate':                     \n",
        "                             '8.8e-09', 'num_tokens': '4.587e+05', 'completions/mean_length':                      \n",
-       "                             '27.25', 'completions/min_length': '21', 'completions/max_length':                    \n",
+       "                             '27.24', 'completions/min_length': '21', 'completions/max_length':                    \n",
        "                             '44', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '27.25',                                        \n",
+       "                             'completions/mean_terminated_length': '27.24',                                        \n",
        "                             'completions/min_terminated_length': '21',                                            \n",
        "                             'completions/max_terminated_length': '44',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.75',                                              \n",
-       "                             'rewards/tool_call_reward/std': '0.4347',                                             \n",
-       "                             'rewards/format_reward/mean': '0.8979',                                               \n",
-       "                             'rewards/format_reward/std': '0.3049', 'reward': '1.199',                             \n",
-       "                             'reward_std': '0.5379', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.0001013', 'entropy': '0.01624', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'rewards/tool_call_reward/mean': '0.7344',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4434',                                             \n",
+       "                             'rewards/format_reward/mean': '0.8822',                                               \n",
+       "                             'rewards/format_reward/std': '0.3246', 'reward': '1.175',                             \n",
+       "                             'reward_std': '0.5569', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.0001488', 'entropy': '0.01623', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '13.17', 'epoch': '0.056'}                                               \n",
+       "                             'step_time': '12.26', 'epoch': '0.056'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:32:01]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027555;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027556;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m12\u001b[0m%|█▏        | \u001b[1m6\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:30<10:14, \u001b[1m13.\u001b[0m97s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m14\u001b[0m%|█▍        |   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m7\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:43<09:49, \u001b[1m13.\u001b[0m71s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.000197', 'grad_norm': '5.104', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:04:20]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029259;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029260;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m12\u001b[0m%|█▏        | \u001b[1m6\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:29<09:56, \u001b[1m13.\u001b[0m55s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m14\u001b[0m%|█▍        |   \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m7\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:41<09:24, \u001b[1m13.\u001b[0m13s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.0003451', 'grad_norm': '5.135', 'learning_rate':   \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '8.8e-09', 'num_tokens': '4.587e+05', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '27.25', 'completions/min_length': '21', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '27.24', 'completions/min_length': '21', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '44', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '27.25',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '27.24',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '21',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '44',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.75',                            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4347',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8979',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.3049', 'reward': '1.199',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5379', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0001013', 'entropy': '0.01624', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7344',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4434',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8822',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.3246', 'reward': '1.175',           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.5569', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.0001488', 'entropy': '0.01623', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '13.17', 'epoch': '0.056'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '12.26', 'epoch': '0.056'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -4225,49 +4214,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:32:18] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             14%|█▍        | 7/50 [01:43<09:49, 13.71s/it]#015 16%|█▌        |                     \n",
-       "                             8/50 [01:58<09:50, 14.06s/it]#015                                                     \n",
-       "                             #015{'loss': '0.005951', 'grad_norm': '1.704', 'learning_rate':                       \n",
+       "
[06/26/26 11:04:31] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             14%|█▍        | 7/50 [01:41<09:24, 13.13s/it]#015 16%|█▌        |                     \n",
+       "                             8/50 [01:56<09:34, 13.69s/it]#015                                                     \n",
+       "                             #015{'loss': '0.006165', 'grad_norm': '1.91', 'learning_rate':                        \n",
        "                             '8.6e-09', 'num_tokens': '5.259e+05', 'completions/mean_length':                      \n",
-       "                             '38.98', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '38.94', 'completions/min_length': '20', 'completions/max_length':                    \n",
        "                             '109', 'completions/clipped_ratio': '0',                                              \n",
-       "                             'completions/mean_terminated_length': '38.98',                                        \n",
+       "                             'completions/mean_terminated_length': '38.94',                                        \n",
        "                             'completions/min_terminated_length': '20',                                            \n",
        "                             'completions/max_terminated_length': '109',                                           \n",
        "                             'rewards/tool_call_reward/mean': '0.6797',                                            \n",
        "                             'rewards/tool_call_reward/std': '0.4684',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9607',                                               \n",
-       "                             'rewards/format_reward/std': '0.1959', 'reward': '1.16',                              \n",
-       "                             'reward_std': '0.5059', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '8.511e-05', 'entropy': '0.01835', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'rewards/format_reward/mean': '0.9685',                                               \n",
+       "                             'rewards/format_reward/std': '0.176', 'reward': '1.164',                              \n",
+       "                             'reward_std': '0.4987', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '5.315e-05', 'entropy': '0.01854', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '14.8', 'epoch': '0.064'}                                                \n",
+       "                             'step_time': '14.87', 'epoch': '0.064'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:32:18]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027563;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027564;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m14\u001b[0m%|█▍        | \u001b[1m7\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:43<09:49, \u001b[1m13.\u001b[0m71s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m16\u001b[0m%|█▌        |   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m8\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:58<09:50, \u001b[1m14.\u001b[0m06s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.005951', 'grad_norm': '1.704', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:04:31]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029267;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029268;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m14\u001b[0m%|█▍        | \u001b[1m7\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:41<09:24, \u001b[1m13.\u001b[0m13s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m16\u001b[0m%|█▌        |   \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m8\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:56<09:34, \u001b[1m13.\u001b[0m69s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.006165', 'grad_norm': '1.91', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '8.6e-09', 'num_tokens': '5.259e+05', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '38.98', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '38.94', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '109', 'completions/clipped_ratio': '0',                            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '38.98',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '38.94',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '109',                         \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.6797',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4684',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9607',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.1959', 'reward': '1.16',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5059', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '8.511e-05', 'entropy': '0.01835', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9685',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.176', 'reward': '1.164',            \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.4987', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '5.315e-05', 'entropy': '0.01854', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '14.8', 'epoch': '0.064'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '14.87', 'epoch': '0.064'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -4276,49 +4265,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:32:29] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             16%|█▌        | 8/50 [01:58<09:50, 14.06s/it]#015 18%|█▊        |                     \n",
-       "                             9/50 [02:10<09:05, 13.31s/it]#015                                                     \n",
-       "                             #015{'loss': '0.001581', 'grad_norm': '2.503', 'learning_rate':                       \n",
+       "
[06/26/26 11:04:47] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             16%|█▌        | 8/50 [01:56<09:34, 13.69s/it]#015 18%|█▊        |                     \n",
+       "                             9/50 [02:08<08:57, 13.10s/it]#015                                                     \n",
+       "                             #015{'loss': '0.001617', 'grad_norm': '2.566', 'learning_rate':                       \n",
        "                             '8.4e-09', 'num_tokens': '5.847e+05', 'completions/mean_length':                      \n",
        "                             '32.96', 'completions/min_length': '24', 'completions/max_length':                    \n",
        "                             '51', 'completions/clipped_ratio': '0',                                               \n",
        "                             'completions/mean_terminated_length': '32.96',                                        \n",
        "                             'completions/min_terminated_length': '24',                                            \n",
        "                             'completions/max_terminated_length': '51',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7891',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4096',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9136',                                               \n",
-       "                             'rewards/format_reward/std': '0.2828', 'reward': '1.246',                             \n",
-       "                             'reward_std': '0.5064', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '7.845e-06', 'entropy': '0.01171', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'rewards/tool_call_reward/mean': '0.7969',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4039',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9215',                                               \n",
+       "                             'rewards/format_reward/std': '0.2708', 'reward': '1.258',                             \n",
+       "                             'reward_std': '0.4945', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '-1.58e-05', 'entropy': '0.01142', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '11.65', 'epoch': '0.072'}                                               \n",
+       "                             'step_time': '11.8', 'epoch': '0.072'}                                                \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:32:29]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027571;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027572;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m16\u001b[0m%|█▌        | \u001b[1m8\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:58<09:50, \u001b[1m14.\u001b[0m06s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m18\u001b[0m%|█▊        |   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m9\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:10<09:05, \u001b[1m13.\u001b[0m31s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.001581', 'grad_norm': '2.503', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:04:47]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029275;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029276;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m16\u001b[0m%|█▌        | \u001b[1m8\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:56<09:34, \u001b[1m13.\u001b[0m69s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m18\u001b[0m%|█▊        |   \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m9\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:08<08:57, \u001b[1m13.\u001b[0m10s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.001617', 'grad_norm': '2.566', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '8.4e-09', 'num_tokens': '5.847e+05', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '32.96', 'completions/min_length': '24', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '51', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '32.96',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '24',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '51',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7891',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4096',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9136',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2828', 'reward': '1.246',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5064', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '7.845e-06', 'entropy': '0.01171', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7969',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4039',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9215',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2708', 'reward': '1.258',           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.4945', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '-1.58e-05', 'entropy': '0.01142', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '11.65', 'epoch': '0.072'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '11.8', 'epoch': '0.072'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -4327,12 +4316,12 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:32:45] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             18%|█▊        | 9/50 [02:10<09:05, 13.31s/it]#015 20%|██        |                     \n",
-       "                             10/50 [02:25<09:19, 13.99s/it]#015                                                    \n",
-       "                             #015{'loss': '-0.003787', 'grad_norm': '3.198', 'learning_rate':                      \n",
-       "                             '8.2e-09', 'num_tokens': '6.406e+05', 'completions/mean_length':                      \n",
+       "
[06/26/26 11:04:58] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             18%|█▊        | 9/50 [02:08<08:57, 13.10s/it]#015 20%|██        |                     \n",
+       "                             10/50 [02:22<09:01, 13.54s/it]#015                                                    \n",
+       "                             #015{'loss': '-0.003781', 'grad_norm': '3.136', 'learning_rate':                      \n",
+       "                             '8.2e-09', 'num_tokens': '6.407e+05', 'completions/mean_length':                      \n",
        "                             '30.45', 'completions/min_length': '19', 'completions/max_length':                    \n",
        "                             '56', 'completions/clipped_ratio': '0',                                               \n",
        "                             'completions/mean_terminated_length': '30.45',                                        \n",
@@ -4343,19 +4332,19 @@
        "                             'rewards/format_reward/mean': '0.9293',                                               \n",
        "                             'rewards/format_reward/std': '0.2581', 'reward': '1.183',                             \n",
        "                             'reward_std': '0.5212', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '-1.742e-05', 'entropy': '0.0144', 'clip_ratio/low_mean': '0',                        \n",
+       "                             '5.048e-05', 'entropy': '0.01439', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '15.51', 'epoch': '0.08'}                                                \n",
+       "                             'step_time': '14.52', 'epoch': '0.08'}                                                \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:32:45]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027579;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027580;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m18\u001b[0m%|█▊        | \u001b[1m9\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:10<09:05, \u001b[1m13.\u001b[0m31s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m20\u001b[0m%|██        |   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m10\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:25<09:19, \u001b[1m13.\u001b[0m99s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.003787', 'grad_norm': '3.198', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '8.2e-09', 'num_tokens': '6.406e+05', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:04:58]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029283;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029284;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m18\u001b[0m%|█▊        | \u001b[1m9\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:08<08:57, \u001b[1m13.\u001b[0m10s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m20\u001b[0m%|██        |   \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m10\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:22<09:01, \u001b[1m13.\u001b[0m54s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.003781', 'grad_norm': '3.136', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '8.2e-09', 'num_tokens': '6.407e+05', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '30.45', 'completions/min_length': '19', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '56', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '30.45',                      \u001b[2m                  \u001b[0m\n",
@@ -4366,10 +4355,10 @@
        "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9293',                             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2581', 'reward': '1.183',           \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'reward_std': '0.5212', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '-1.742e-05', 'entropy': '0.0144', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '5.048e-05', 'entropy': '0.01439', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '15.51', 'epoch': '0.08'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '14.52', 'epoch': '0.08'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -4378,15 +4367,15 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:32:55] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             20%|██        | 10/50 [02:25<09:19, 13.99s/it]#015 22%|██▏       |                    \n",
-       "                             11/50 [02:37<08:38, 13.30s/it]#015                                                    \n",
-       "                             #015{'loss': '0.001828', 'grad_norm': '1.217', 'learning_rate':                       \n",
+       "
[06/26/26 11:05:14] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             20%|██        | 10/50 [02:22<09:01, 13.54s/it]#015 22%|██▏       |                    \n",
+       "                             11/50 [02:35<08:38, 13.31s/it]#015                                                    \n",
+       "                             #015{'loss': '0.00183', 'grad_norm': '1.218', 'learning_rate':                        \n",
        "                             '8e-09', 'num_tokens': '6.995e+05', 'completions/mean_length':                        \n",
-       "                             '31.66', 'completions/min_length': '23', 'completions/max_length':                    \n",
+       "                             '31.61', 'completions/min_length': '23', 'completions/max_length':                    \n",
        "                             '47', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '31.66',                                        \n",
+       "                             'completions/mean_terminated_length': '31.61',                                        \n",
        "                             'completions/min_terminated_length': '23',                                            \n",
        "                             'completions/max_terminated_length': '47',                                            \n",
        "                             'rewards/tool_call_reward/mean': '0.8047',                                            \n",
@@ -4394,22 +4383,22 @@
        "                             'rewards/format_reward/mean': '0.9372',                                               \n",
        "                             'rewards/format_reward/std': '0.2442', 'reward': '1.273',                             \n",
        "                             'reward_std': '0.4736', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '3.692e-05', 'entropy': '0.01143', 'clip_ratio/low_mean': '0',                        \n",
+       "                             '3.313e-05', 'entropy': '0.01158', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '11.71', 'epoch': '0.088'}                                               \n",
+       "                             'step_time': '12.76', 'epoch': '0.088'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:32:55]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027587;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027588;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m20\u001b[0m%|██        | \u001b[1m10\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:25<09:19, \u001b[1m13.\u001b[0m99s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m22\u001b[0m%|██▏       |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m11\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:37<08:38, \u001b[1m13.\u001b[0m30s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.001828', 'grad_norm': '1.217', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:05:14]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029291;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029292;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m20\u001b[0m%|██        | \u001b[1m10\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:22<09:01, \u001b[1m13.\u001b[0m54s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m22\u001b[0m%|██▏       |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m11\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:35<08:38, \u001b[1m13.\u001b[0m31s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.00183', 'grad_norm': '1.218', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '8e-09', 'num_tokens': '6.995e+05', 'completions/mean_length':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '31.66', 'completions/min_length': '23', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '31.61', 'completions/min_length': '23', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '47', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '31.66',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '31.61',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '23',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '47',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8047',                          \u001b[2m                  \u001b[0m\n",
@@ -4417,10 +4406,10 @@
        "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9372',                             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2442', 'reward': '1.273',           \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'reward_std': '0.4736', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '3.692e-05', 'entropy': '0.01143', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '3.313e-05', 'entropy': '0.01158', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '11.71', 'epoch': '0.088'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '12.76', 'epoch': '0.088'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -4429,49 +4418,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:33:06] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             22%|██▏       | 11/50 [02:37<08:38, 13.30s/it]#015 24%|██▍       |                    \n",
-       "                             12/50 [02:48<08:03, 12.73s/it]#015                                                    \n",
-       "                             #015{'loss': '-0.003021', 'grad_norm': '2.149', 'learning_rate':                      \n",
+       "
[06/26/26 11:05:26] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             22%|██▏       | 11/50 [02:35<08:38, 13.31s/it]#015 24%|██▍       |                    \n",
+       "                             12/50 [02:47<08:16, 13.07s/it]#015                                                    \n",
+       "                             #015{'loss': '-0.002743', 'grad_norm': '2.056', 'learning_rate':                      \n",
        "                             '7.8e-09', 'num_tokens': '7.648e+05', 'completions/mean_length':                      \n",
        "                             '31.52', 'completions/min_length': '22', 'completions/max_length':                    \n",
        "                             '40', 'completions/clipped_ratio': '0',                                               \n",
        "                             'completions/mean_terminated_length': '31.52',                                        \n",
        "                             'completions/min_terminated_length': '22',                                            \n",
        "                             'completions/max_terminated_length': '40',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7266',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4475',                                             \n",
-       "                             'rewards/format_reward/mean': '0.8979',                                               \n",
-       "                             'rewards/format_reward/std': '0.305', 'reward': '1.175',                              \n",
-       "                             'reward_std': '0.5461', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.000162', 'entropy': '0.02277', 'clip_ratio/low_mean': '0',                         \n",
+       "                             'rewards/tool_call_reward/mean': '0.7344',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4434',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9136',                                               \n",
+       "                             'rewards/format_reward/std': '0.283', 'reward': '1.191',                              \n",
+       "                             'reward_std': '0.5297', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '6.099e-05', 'entropy': '0.02272', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '11.42', 'epoch': '0.096'}                                               \n",
+       "                             'step_time': '12.53', 'epoch': '0.096'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:33:06]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027595;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027596;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m22\u001b[0m%|██▏       | \u001b[1m11\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:37<08:38, \u001b[1m13.\u001b[0m30s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m24\u001b[0m%|██▍       |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m12\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:48<08:03, \u001b[1m12.\u001b[0m73s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.003021', 'grad_norm': '2.149', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:05:26]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029299;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029300;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m22\u001b[0m%|██▏       | \u001b[1m11\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:35<08:38, \u001b[1m13.\u001b[0m31s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m24\u001b[0m%|██▍       |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m12\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:47<08:16, \u001b[1m13.\u001b[0m07s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.002743', 'grad_norm': '2.056', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '7.8e-09', 'num_tokens': '7.648e+05', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '31.52', 'completions/min_length': '22', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '40', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '31.52',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '22',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '40',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7266',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4475',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8979',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.305', 'reward': '1.175',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5461', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.000162', 'entropy': '0.02277', 'clip_ratio/low_mean': '0',       \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7344',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4434',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9136',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.283', 'reward': '1.191',            \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.5297', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '6.099e-05', 'entropy': '0.02272', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '11.42', 'epoch': '0.096'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '12.53', 'epoch': '0.096'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -4480,49 +4469,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:33:23] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             24%|██▍       | 12/50 [02:48<08:03, 12.73s/it]#015 26%|██▌       |                    \n",
-       "                             13/50 [03:04<08:25, 13.67s/it]#015                                                    \n",
-       "                             #015{'loss': '0.001063', 'grad_norm': '1.959', 'learning_rate':                       \n",
+       "
[06/26/26 11:05:42] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             24%|██▍       | 12/50 [02:47<08:16, 13.07s/it]#015 26%|██▌       |                    \n",
+       "                             13/50 [03:05<08:48, 14.29s/it]#015                                                    \n",
+       "                             #015{'loss': '0.001483', 'grad_norm': '1.707', 'learning_rate':                       \n",
        "                             '7.6e-09', 'num_tokens': '8.236e+05', 'completions/mean_length':                      \n",
-       "                             '32.22', 'completions/min_length': '21', 'completions/max_length':                    \n",
+       "                             '32.16', 'completions/min_length': '21', 'completions/max_length':                    \n",
        "                             '59', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '32.22',                                        \n",
+       "                             'completions/mean_terminated_length': '32.16',                                        \n",
        "                             'completions/min_terminated_length': '21',                                            \n",
        "                             'completions/max_terminated_length': '59',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7188',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4514',                                             \n",
+       "                             'rewards/tool_call_reward/mean': '0.7109',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4551',                                             \n",
        "                             'rewards/format_reward/mean': '0.9607',                                               \n",
-       "                             'rewards/format_reward/std': '0.1956', 'reward': '1.199',                             \n",
-       "                             'reward_std': '0.4917', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.000176', 'entropy': '0.0192', 'clip_ratio/low_mean': '0',                          \n",
+       "                             'rewards/format_reward/std': '0.1956', 'reward': '1.191',                             \n",
+       "                             'reward_std': '0.4948', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.0001074', 'entropy': '0.01953', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '15.83', 'epoch': '0.104'}                                               \n",
+       "                             'step_time': '17.07', 'epoch': '0.104'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:33:23]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027603;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027604;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m24\u001b[0m%|██▍       | \u001b[1m12\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:48<08:03, \u001b[1m12.\u001b[0m73s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m26\u001b[0m%|██▌       |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m13\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:04<08:25, \u001b[1m13.\u001b[0m67s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.001063', 'grad_norm': '1.959', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:05:42]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029307;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029308;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m24\u001b[0m%|██▍       | \u001b[1m12\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:47<08:16, \u001b[1m13.\u001b[0m07s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m26\u001b[0m%|██▌       |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m13\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:05<08:48, \u001b[1m14.\u001b[0m29s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.001483', 'grad_norm': '1.707', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '7.6e-09', 'num_tokens': '8.236e+05', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '32.22', 'completions/min_length': '21', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '32.16', 'completions/min_length': '21', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '59', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '32.22',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '32.16',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '21',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '59',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7188',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4514',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7109',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4551',                           \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9607',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.1956', 'reward': '1.199',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.4917', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.000176', 'entropy': '0.0192', 'clip_ratio/low_mean': '0',        \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.1956', 'reward': '1.191',           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.4948', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.0001074', 'entropy': '0.01953', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '15.83', 'epoch': '0.104'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '17.07', 'epoch': '0.104'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -4531,49 +4520,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:33:33] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             26%|██▌       | 13/50 [03:04<08:25, 13.67s/it]#015 28%|██▊       |                    \n",
-       "                             14/50 [03:17<08:04, 13.46s/it]#015                                                    \n",
-       "                             #015{'loss': '0.02065', 'grad_norm': '5.698', 'learning_rate':                        \n",
+       "
[06/26/26 11:05:53] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             26%|██▌       | 13/50 [03:05<08:48, 14.29s/it]#015 28%|██▊       |                    \n",
+       "                             14/50 [03:19<08:32, 14.24s/it]#015                                                    \n",
+       "                             #015{'loss': '0.02113', 'grad_norm': '5.839', 'learning_rate':                        \n",
        "                             '7.4e-09', 'num_tokens': '8.798e+05', 'completions/mean_length':                      \n",
-       "                             '35.79', 'completions/min_length': '15', 'completions/max_length':                    \n",
+       "                             '35.77', 'completions/min_length': '15', 'completions/max_length':                    \n",
        "                             '73', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '35.79',                                        \n",
+       "                             'completions/mean_terminated_length': '35.77',                                        \n",
        "                             'completions/min_terminated_length': '15',                                            \n",
        "                             'completions/max_terminated_length': '73',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7969',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4039',                                             \n",
-       "                             'rewards/format_reward/mean': '0.937', 'rewards/format_reward/std':                   \n",
-       "                             '0.245', 'reward': '1.265', 'reward_std': '0.4783',                                   \n",
-       "                             'frac_reward_zero_std': '0', 'kl': '0.0001055', 'entropy':                            \n",
-       "                             '0.02258', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',                     \n",
-       "                             'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',                              \n",
-       "                             'clip_ratio/region_mean': '0', 'step_time': '12.96', 'epoch':                         \n",
-       "                             '0.112'}                                                                              \n",
+       "                             'rewards/tool_call_reward/mean': '0.8047',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.398',                                              \n",
+       "                             'rewards/format_reward/mean': '0.9448',                                               \n",
+       "                             'rewards/format_reward/std': '0.2302', 'reward': '1.277',                             \n",
+       "                             'reward_std': '0.4652', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '9.532e-05', 'entropy': '0.02313', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '14.13', 'epoch': '0.112'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:33:33]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027611;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027612;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m26\u001b[0m%|██▌       | \u001b[1m13\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:04<08:25, \u001b[1m13.\u001b[0m67s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m28\u001b[0m%|██▊       |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m14\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:17<08:04, \u001b[1m13.\u001b[0m46s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.02065', 'grad_norm': '5.698', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:05:53]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029315;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029316;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m26\u001b[0m%|██▌       | \u001b[1m13\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:05<08:48, \u001b[1m14.\u001b[0m29s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m28\u001b[0m%|██▊       |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m14\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:19<08:32, \u001b[1m14.\u001b[0m24s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.02113', 'grad_norm': '5.839', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '7.4e-09', 'num_tokens': '8.798e+05', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '35.79', 'completions/min_length': '15', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '35.77', 'completions/min_length': '15', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '73', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '35.79',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '35.77',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '15',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '73',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7969',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4039',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.937', 'rewards/format_reward/std': \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.245', 'reward': '1.265', 'reward_std': '0.4783',                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'frac_reward_zero_std': '0', 'kl': '0.0001055', 'entropy':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.02258', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/region_mean': '0', 'step_time': '12.96', 'epoch':       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.112'\u001b[1m}\u001b[0m                                                            \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8047',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.398',                            \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9448',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2302', 'reward': '1.277',           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.4652', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '9.532e-05', 'entropy': '0.02313', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'step_time': '14.13', 'epoch': '0.112'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -4582,49 +4571,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:33:49] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             28%|██▊       | 14/50 [03:17<08:04, 13.46s/it]#015 30%|███       |                    \n",
-       "                             15/50 [03:33<08:15, 14.15s/it]#015                                                    \n",
-       "                             #015{'loss': '0.003047', 'grad_norm': '1.325', 'learning_rate':                       \n",
+       "
[06/26/26 11:06:15] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             28%|██▊       | 14/50 [03:19<08:32, 14.24s/it]#015 30%|███       |                    \n",
+       "                             15/50 [03:36<08:51, 15.17s/it]#015                                                    \n",
+       "                             #015{'loss': '0.001838', 'grad_norm': '1.245', 'learning_rate':                       \n",
        "                             '7.2e-09', 'num_tokens': '9.451e+05', 'completions/mean_length':                      \n",
-       "                             '42.41', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '42.38', 'completions/min_length': '20', 'completions/max_length':                    \n",
        "                             '128', 'completions/clipped_ratio': '0.0625',                                         \n",
-       "                             'completions/mean_terminated_length': '36.7',                                         \n",
+       "                             'completions/mean_terminated_length': '36.67',                                        \n",
        "                             'completions/min_terminated_length': '20',                                            \n",
        "                             'completions/max_terminated_length': '62',                                            \n",
        "                             'rewards/tool_call_reward/mean': '0.8359',                                            \n",
        "                             'rewards/tool_call_reward/std': '0.3718',                                             \n",
-       "                             'rewards/format_reward/mean': '0.8505',                                               \n",
-       "                             'rewards/format_reward/std': '0.3594', 'reward': '1.261',                             \n",
-       "                             'reward_std': '0.5445', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '3.676e-05', 'entropy': '0.009485', 'clip_ratio/low_mean': '0',                       \n",
+       "                             'rewards/format_reward/mean': '0.8427',                                               \n",
+       "                             'rewards/format_reward/std': '0.367', 'reward': '1.257',                              \n",
+       "                             'reward_std': '0.5518', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '2.964e-05', 'entropy': '0.009179', 'clip_ratio/low_mean': '0',                       \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '15.74', 'epoch': '0.12'}                                                \n",
+       "                             'step_time': '17.31', 'epoch': '0.12'}                                                \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:33:49]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027619;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027620;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m28\u001b[0m%|██▊       | \u001b[1m14\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:17<08:04, \u001b[1m13.\u001b[0m46s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m30\u001b[0m%|███       |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m15\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:33<08:15, \u001b[1m14.\u001b[0m15s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.003047', 'grad_norm': '1.325', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:06:15]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029323;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029324;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m28\u001b[0m%|██▊       | \u001b[1m14\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:19<08:32, \u001b[1m14.\u001b[0m24s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m30\u001b[0m%|███       |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m15\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:36<08:51, \u001b[1m15.\u001b[0m17s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.001838', 'grad_norm': '1.245', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '7.2e-09', 'num_tokens': '9.451e+05', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '42.41', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '42.38', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '128', 'completions/clipped_ratio': '0.0625',                       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '36.7',                       \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '36.67',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '62',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8359',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.3718',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8505',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.3594', 'reward': '1.261',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5445', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '3.676e-05', 'entropy': '0.009485', 'clip_ratio/low_mean': '0',     \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8427',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.367', 'reward': '1.257',            \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.5518', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '2.964e-05', 'entropy': '0.009179', 'clip_ratio/low_mean': '0',     \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '15.74', 'epoch': '0.12'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '17.31', 'epoch': '0.12'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -4633,49 +4622,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:34:05] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             30%|███       | 15/50 [03:33<08:15, 14.15s/it]#015 32%|███▏      |                    \n",
-       "                             16/50 [03:49<08:20, 14.72s/it]#015                                                    \n",
-       "                             #015{'loss': '-0.001221', 'grad_norm': '2.042', 'learning_rate':                      \n",
+       "
[06/26/26 11:06:31] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             30%|███       | 15/50 [03:36<08:51, 15.17s/it]#015 32%|███▏      |                    \n",
+       "                             16/50 [03:51<08:38, 15.25s/it]#015                                                    \n",
+       "                             #015{'loss': '-0.001349', 'grad_norm': '1.989', 'learning_rate':                      \n",
        "                             '7e-09', 'num_tokens': '1.022e+06', 'completions/mean_length':                        \n",
-       "                             '32.47', 'completions/min_length': '16', 'completions/max_length':                    \n",
+       "                             '32.54', 'completions/min_length': '16', 'completions/max_length':                    \n",
        "                             '57', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '32.47',                                        \n",
+       "                             'completions/mean_terminated_length': '32.54',                                        \n",
        "                             'completions/min_terminated_length': '16',                                            \n",
        "                             'completions/max_terminated_length': '57',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7656',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4253',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9057',                                               \n",
-       "                             'rewards/format_reward/std': '0.2944', 'reward': '1.218',                             \n",
-       "                             'reward_std': '0.5247', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '5.935e-05', 'entropy': '0.02726', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'rewards/tool_call_reward/mean': '0.7734',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4203',                                             \n",
+       "                             'rewards/format_reward/mean': '0.8978',                                               \n",
+       "                             'rewards/format_reward/std': '0.305', 'reward': '1.222',                              \n",
+       "                             'reward_std': '0.5287', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '8.802e-05', 'entropy': '0.02825', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '16.01', 'epoch': '0.128'}                                               \n",
+       "                             'step_time': '15.43', 'epoch': '0.128'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:34:05]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027627;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027628;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m30\u001b[0m%|███       | \u001b[1m15\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:33<08:15, \u001b[1m14.\u001b[0m15s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m32\u001b[0m%|███▏      |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m16\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:49<08:20, \u001b[1m14.\u001b[0m72s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.001221', 'grad_norm': '2.042', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:06:31]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029331;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029332;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m30\u001b[0m%|███       | \u001b[1m15\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:36<08:51, \u001b[1m15.\u001b[0m17s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m32\u001b[0m%|███▏      |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m16\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:51<08:38, \u001b[1m15.\u001b[0m25s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.001349', 'grad_norm': '1.989', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '7e-09', 'num_tokens': '1.022e+06', 'completions/mean_length':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '32.47', 'completions/min_length': '16', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '32.54', 'completions/min_length': '16', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '57', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '32.47',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '32.54',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '16',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '57',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7656',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4253',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9057',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2944', 'reward': '1.218',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5247', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '5.935e-05', 'entropy': '0.02726', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7734',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4203',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8978',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.305', 'reward': '1.222',            \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.5287', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '8.802e-05', 'entropy': '0.02825', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '16.01', 'epoch': '0.128'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '15.43', 'epoch': '0.128'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -4684,49 +4673,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:34:22] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             32%|███▏      | 16/50 [03:49<08:20, 14.72s/it]#015 34%|███▍      |                    \n",
-       "                             17/50 [04:05<08:14, 14.98s/it]#015                                                    \n",
-       "                             #015{'loss': '0.007587', 'grad_norm': '2.109', 'learning_rate':                       \n",
+       "
[06/26/26 11:06:41] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             32%|███▏      | 16/50 [03:51<08:38, 15.25s/it]#015 34%|███▍      |                    \n",
+       "                             17/50 [04:07<08:27, 15.38s/it]#015                                                    \n",
+       "                             #015{'loss': '0.0009151', 'grad_norm': '1.993', 'learning_rate':                      \n",
        "                             '6.8e-09', 'num_tokens': '1.075e+06', 'completions/mean_length':                      \n",
-       "                             '36.02', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '35.89', 'completions/min_length': '20', 'completions/max_length':                    \n",
        "                             '128', 'completions/clipped_ratio': '0.0625',                                         \n",
-       "                             'completions/mean_terminated_length': '29.88',                                        \n",
+       "                             'completions/mean_terminated_length': '29.75',                                        \n",
        "                             'completions/min_terminated_length': '20',                                            \n",
-       "                             'completions/max_terminated_length': '53',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.8594',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.349',                                              \n",
-       "                             'rewards/format_reward/mean': '0.8819',                                               \n",
-       "                             'rewards/format_reward/std': '0.3253', 'reward': '1.3',                               \n",
-       "                             'reward_std': '0.5005', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '3.856e-06', 'entropy': '0.0109', 'clip_ratio/low_mean': '0',                         \n",
+       "                             'completions/max_terminated_length': '45',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.8516',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.3569',                                             \n",
+       "                             'rewards/format_reward/mean': '0.8663',                                               \n",
+       "                             'rewards/format_reward/std': '0.3431', 'reward': '1.285',                             \n",
+       "                             'reward_std': '0.5211', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '1.607e-05', 'entropy': '0.009799', 'clip_ratio/low_mean': '0',                       \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '15.59', 'epoch': '0.136'}                                               \n",
+       "                             'step_time': '15.67', 'epoch': '0.136'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:34:22]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027635;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027636;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m32\u001b[0m%|███▏      | \u001b[1m16\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:49<08:20, \u001b[1m14.\u001b[0m72s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m34\u001b[0m%|███▍      |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m17\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:05<08:14, \u001b[1m14.\u001b[0m98s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.007587', 'grad_norm': '2.109', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:06:41]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029339;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029340;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m32\u001b[0m%|███▏      | \u001b[1m16\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:51<08:38, \u001b[1m15.\u001b[0m25s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m34\u001b[0m%|███▍      |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m17\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:07<08:27, \u001b[1m15.\u001b[0m38s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.0009151', 'grad_norm': '1.993', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '6.8e-09', 'num_tokens': '1.075e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '36.02', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '35.89', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '128', 'completions/clipped_ratio': '0.0625',                       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '29.88',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '29.75',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '53',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8594',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.349',                            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8819',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.3253', 'reward': '1.3',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5005', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '3.856e-06', 'entropy': '0.0109', 'clip_ratio/low_mean': '0',       \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '45',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8516',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.3569',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8663',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.3431', 'reward': '1.285',           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.5211', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '1.607e-05', 'entropy': '0.009799', 'clip_ratio/low_mean': '0',     \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '15.59', 'epoch': '0.136'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '15.67', 'epoch': '0.136'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -4735,49 +4724,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:34:32] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             34%|███▍      | 17/50 [04:05<08:14, 14.98s/it]#015 36%|███▌      |                    \n",
-       "                             18/50 [04:17<07:35, 14.23s/it]#015                                                    \n",
-       "                             #015{'loss': '0.004968', 'grad_norm': '2.632', 'learning_rate':                       \n",
+       "
[06/26/26 11:06:57] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             34%|███▍      | 17/50 [04:07<08:27, 15.38s/it]#015 36%|███▌      |                    \n",
+       "                             18/50 [04:20<07:46, 14.57s/it]#015                                                    \n",
+       "                             #015{'loss': '0.00337', 'grad_norm': '1.611', 'learning_rate':                        \n",
        "                             '6.6e-09', 'num_tokens': '1.14e+06', 'completions/mean_length':                       \n",
-       "                             '33.13', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '33.16', 'completions/min_length': '20', 'completions/max_length':                    \n",
        "                             '65', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '33.13',                                        \n",
+       "                             'completions/mean_terminated_length': '33.16',                                        \n",
        "                             'completions/min_terminated_length': '20',                                            \n",
        "                             'completions/max_terminated_length': '65',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.8125',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.3918',                                             \n",
+       "                             'rewards/tool_call_reward/mean': '0.7656',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4253',                                             \n",
        "                             'rewards/format_reward/mean': '0.9293',                                               \n",
-       "                             'rewards/format_reward/std': '0.2581', 'reward': '1.277',                             \n",
-       "                             'reward_std': '0.4776', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '8.186e-05', 'entropy': '0.01777', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'rewards/format_reward/std': '0.2581', 'reward': '1.23',                              \n",
+       "                             'reward_std': '0.5021', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '5.279e-05', 'entropy': '0.02029', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '12.48', 'epoch': '0.144'}                                               \n",
+       "                             'step_time': '12.65', 'epoch': '0.144'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:34:32]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027643;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027644;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m34\u001b[0m%|███▍      | \u001b[1m17\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:05<08:14, \u001b[1m14.\u001b[0m98s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m36\u001b[0m%|███▌      |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m18\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:17<07:35, \u001b[1m14.\u001b[0m23s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.004968', 'grad_norm': '2.632', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:06:57]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029347;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029348;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m34\u001b[0m%|███▍      | \u001b[1m17\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:07<08:27, \u001b[1m15.\u001b[0m38s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m36\u001b[0m%|███▌      |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m18\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:20<07:46, \u001b[1m14.\u001b[0m57s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.00337', 'grad_norm': '1.611', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '6.6e-09', 'num_tokens': '1.14e+06', 'completions/mean_length':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '33.13', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '33.16', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '65', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '33.13',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '33.16',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '65',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8125',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.3918',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7656',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4253',                           \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9293',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2581', 'reward': '1.277',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.4776', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '8.186e-05', 'entropy': '0.01777', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2581', 'reward': '1.23',            \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.5021', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '5.279e-05', 'entropy': '0.02029', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '12.48', 'epoch': '0.144'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '12.65', 'epoch': '0.144'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -4786,49 +4775,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:34:48] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             36%|███▌      | 18/50 [04:17<07:35, 14.23s/it]#015 38%|███▊      |                    \n",
-       "                             19/50 [04:33<07:34, 14.65s/it]#015                                                    \n",
-       "                             #015{'loss': '-0.008127', 'grad_norm': '2.291', 'learning_rate':                      \n",
+       "
[06/26/26 11:07:08] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             36%|███▌      | 18/50 [04:20<07:46, 14.57s/it]#015 38%|███▊      |                    \n",
+       "                             19/50 [04:34<07:29, 14.51s/it]#015                                                    \n",
+       "                             #015{'loss': '-0.00885', 'grad_norm': '2.367', 'learning_rate':                       \n",
        "                             '6.4e-09', 'num_tokens': '1.201e+06', 'completions/mean_length':                      \n",
-       "                             '33.46', 'completions/min_length': '16', 'completions/max_length':                    \n",
+       "                             '33.41', 'completions/min_length': '16', 'completions/max_length':                    \n",
        "                             '50', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '33.46',                                        \n",
+       "                             'completions/mean_terminated_length': '33.41',                                        \n",
        "                             'completions/min_terminated_length': '16',                                            \n",
        "                             'completions/max_terminated_length': '50',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.8516',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.3569',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9529',                                               \n",
-       "                             'rewards/format_reward/std': '0.2134', 'reward': '1.328',                             \n",
-       "                             'reward_std': '0.4234', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '5.495e-05', 'entropy': '0.01459', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'rewards/tool_call_reward/mean': '0.8672',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.3407',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9686',                                               \n",
+       "                             'rewards/format_reward/std': '0.1757', 'reward': '1.351',                             \n",
+       "                             'reward_std': '0.3889', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.0001576', 'entropy': '0.01414', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '15.6', 'epoch': '0.152'}                                                \n",
+       "                             'step_time': '14.36', 'epoch': '0.152'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:34:48]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027651;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027652;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m36\u001b[0m%|███▌      | \u001b[1m18\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:17<07:35, \u001b[1m14.\u001b[0m23s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m38\u001b[0m%|███▊      |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m19\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:33<07:34, \u001b[1m14.\u001b[0m65s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.008127', 'grad_norm': '2.291', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:07:08]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029355;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029356;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m36\u001b[0m%|███▌      | \u001b[1m18\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:20<07:46, \u001b[1m14.\u001b[0m57s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m38\u001b[0m%|███▊      |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m19\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:34<07:29, \u001b[1m14.\u001b[0m51s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.00885', 'grad_norm': '2.367', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '6.4e-09', 'num_tokens': '1.201e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '33.46', 'completions/min_length': '16', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '33.41', 'completions/min_length': '16', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '50', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '33.46',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '33.41',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '16',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '50',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8516',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.3569',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9529',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2134', 'reward': '1.328',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.4234', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '5.495e-05', 'entropy': '0.01459', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8672',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.3407',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9686',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.1757', 'reward': '1.351',           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.3889', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.0001576', 'entropy': '0.01414', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '15.6', 'epoch': '0.152'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '14.36', 'epoch': '0.152'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -4837,49 +4826,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:34:59] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             38%|███▊      | 19/50 [04:33<07:34, 14.65s/it]#015 40%|████      |                    \n",
-       "                             20/50 [04:45<06:55, 13.86s/it]#015                                                    \n",
-       "                             #015{'loss': '-0.002019', 'grad_norm': '2.159', 'learning_rate':                      \n",
+       "
[06/26/26 11:07:24] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             38%|███▊      | 19/50 [04:34<07:29, 14.51s/it]#015 40%|████      |                    \n",
+       "                             20/50 [04:46<06:53, 13.79s/it]#015                                                    \n",
+       "                             #015{'loss': '-0.003722', 'grad_norm': '2.091', 'learning_rate':                      \n",
        "                             '6.2e-09', 'num_tokens': '1.272e+06', 'completions/mean_length':                      \n",
-       "                             '32.3', 'completions/min_length': '19', 'completions/max_length':                     \n",
+       "                             '32.25', 'completions/min_length': '19', 'completions/max_length':                    \n",
        "                             '53', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '32.3',                                         \n",
+       "                             'completions/mean_terminated_length': '32.25',                                        \n",
        "                             'completions/min_terminated_length': '19',                                            \n",
        "                             'completions/max_terminated_length': '53',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7266',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4475',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9372',                                               \n",
-       "                             'rewards/format_reward/std': '0.2443', 'reward': '1.195',                             \n",
-       "                             'reward_std': '0.511', 'frac_reward_zero_std': '0', 'kl':                             \n",
-       "                             '0.0001452', 'entropy': '0.03184', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'rewards/tool_call_reward/mean': '0.7344',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4434',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9293',                                               \n",
+       "                             'rewards/format_reward/std': '0.258', 'reward': '1.199',                              \n",
+       "                             'reward_std': '0.5153', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '8.643e-05', 'entropy': '0.03131', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '12.01', 'epoch': '0.16'}                                                \n",
+       "                             'step_time': '12.09', 'epoch': '0.16'}                                                \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:34:59]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027659;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027660;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m38\u001b[0m%|███▊      | \u001b[1m19\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:33<07:34, \u001b[1m14.\u001b[0m65s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m40\u001b[0m%|████      |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m20\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:45<06:55, \u001b[1m13.\u001b[0m86s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.002019', 'grad_norm': '2.159', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:07:24]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029363;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029364;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m38\u001b[0m%|███▊      | \u001b[1m19\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:34<07:29, \u001b[1m14.\u001b[0m51s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m40\u001b[0m%|████      |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m20\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:46<06:53, \u001b[1m13.\u001b[0m79s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.003722', 'grad_norm': '2.091', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '6.2e-09', 'num_tokens': '1.272e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '32.3', 'completions/min_length': '19', 'completions/max_length':   \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '32.25', 'completions/min_length': '19', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '53', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '32.3',                       \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '32.25',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '19',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '53',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7266',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4475',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9372',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2443', 'reward': '1.195',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.511', 'frac_reward_zero_std': '0', 'kl':           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0001452', 'entropy': '0.03184', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7344',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4434',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9293',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.258', 'reward': '1.199',            \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.5153', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '8.643e-05', 'entropy': '0.03131', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '12.01', 'epoch': '0.16'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '12.09', 'epoch': '0.16'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -4888,49 +4877,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:35:15] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             40%|████      | 20/50 [04:45<06:55, 13.86s/it]#015 42%|████▏     |                    \n",
-       "                             21/50 [04:57<06:25, 13.28s/it]#015                                                    \n",
-       "                             #015{'loss': '-0.001169', 'grad_norm': '2.646', 'learning_rate':                      \n",
+       "
[06/26/26 11:07:35] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             40%|████      | 20/50 [04:46<06:53, 13.79s/it]#015 42%|████▏     |                    \n",
+       "                             21/50 [04:58<06:24, 13.25s/it]#015                                                    \n",
+       "                             #015{'loss': '-0.0004618', 'grad_norm': '2.553', 'learning_rate':                     \n",
        "                             '6e-09', 'num_tokens': '1.329e+06', 'completions/mean_length':                        \n",
-       "                             '30.85', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '30.9', 'completions/min_length': '20', 'completions/max_length':                     \n",
        "                             '54', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '30.85',                                        \n",
+       "                             'completions/mean_terminated_length': '30.9',                                         \n",
        "                             'completions/min_terminated_length': '20',                                            \n",
        "                             'completions/max_terminated_length': '54',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7969',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4039',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9058',                                               \n",
-       "                             'rewards/format_reward/std': '0.2941', 'reward': '1.25',                              \n",
-       "                             'reward_std': '0.5103', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.0001261', 'entropy': '0.01297', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'rewards/tool_call_reward/mean': '0.7891',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4096',                                             \n",
+       "                             'rewards/format_reward/mean': '0.8901',                                               \n",
+       "                             'rewards/format_reward/std': '0.315', 'reward': '1.234',                              \n",
+       "                             'reward_std': '0.5291', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.0001838', 'entropy': '0.01313', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '11.92', 'epoch': '0.168'}                                               \n",
+       "                             'step_time': '11.98', 'epoch': '0.168'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:35:15]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027667;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027668;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m40\u001b[0m%|████      | \u001b[1m20\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:45<06:55, \u001b[1m13.\u001b[0m86s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m42\u001b[0m%|████▏     |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m21\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:57<06:25, \u001b[1m13.\u001b[0m28s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.001169', 'grad_norm': '2.646', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:07:35]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029371;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029372;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m40\u001b[0m%|████      | \u001b[1m20\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:46<06:53, \u001b[1m13.\u001b[0m79s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m42\u001b[0m%|████▏     |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m21\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:58<06:24, \u001b[1m13.\u001b[0m25s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.0004618', 'grad_norm': '2.553', 'learning_rate':   \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '6e-09', 'num_tokens': '1.329e+06', 'completions/mean_length':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '30.85', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '30.9', 'completions/min_length': '20', 'completions/max_length':   \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '54', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '30.85',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '30.9',                       \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '54',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7969',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4039',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9058',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2941', 'reward': '1.25',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5103', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0001261', 'entropy': '0.01297', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7891',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4096',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8901',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.315', 'reward': '1.234',            \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.5291', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.0001838', 'entropy': '0.01313', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '11.92', 'epoch': '0.168'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '11.98', 'epoch': '0.168'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -4939,49 +4928,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:35:26] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             42%|████▏     | 21/50 [04:57<06:25, 13.28s/it]#015 44%|████▍     |                    \n",
-       "                             22/50 [05:10<06:14, 13.38s/it]#015                                                    \n",
-       "                             #015{'loss': '0.003743', 'grad_norm': '1.94', 'learning_rate':                        \n",
+       "
[06/26/26 11:07:51] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             42%|████▏     | 21/50 [04:58<06:24, 13.25s/it]#015 44%|████▍     |                    \n",
+       "                             22/50 [05:12<06:17, 13.48s/it]#015                                                    \n",
+       "                             #015{'loss': '0.003621', 'grad_norm': '2.035', 'learning_rate':                       \n",
        "                             '5.8e-09', 'num_tokens': '1.393e+06', 'completions/mean_length':                      \n",
-       "                             '30.69', 'completions/min_length': '18', 'completions/max_length':                    \n",
+       "                             '30.68', 'completions/min_length': '18', 'completions/max_length':                    \n",
        "                             '45', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '30.69',                                        \n",
+       "                             'completions/mean_terminated_length': '30.68',                                        \n",
        "                             'completions/min_terminated_length': '18',                                            \n",
        "                             'completions/max_terminated_length': '45',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7109',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4551',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9607',                                               \n",
-       "                             'rewards/format_reward/std': '0.1956', 'reward': '1.191',                             \n",
-       "                             'reward_std': '0.4948', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '9.293e-05', 'entropy': '0.01876', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '13.59', 'epoch': '0.176'}                                               \n",
+       "                             'rewards/tool_call_reward/mean': '0.7031',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4587',                                             \n",
+       "                             'rewards/format_reward/mean': '0.945', 'rewards/format_reward/std':                   \n",
+       "                             '0.2295', 'reward': '1.176', 'reward_std': '0.5124',                                  \n",
+       "                             'frac_reward_zero_std': '0', 'kl': '0.0001385', 'entropy':                            \n",
+       "                             '0.01903', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',                     \n",
+       "                             'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',                              \n",
+       "                             'clip_ratio/region_mean': '0', 'step_time': '13.99', 'epoch':                         \n",
+       "                             '0.176'}                                                                              \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:35:26]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027675;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027676;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m42\u001b[0m%|████▏     | \u001b[1m21\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:57<06:25, \u001b[1m13.\u001b[0m28s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m44\u001b[0m%|████▍     |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m22\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:10<06:14, \u001b[1m13.\u001b[0m38s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.003743', 'grad_norm': '1.94', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:07:51]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029379;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029380;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m42\u001b[0m%|████▏     | \u001b[1m21\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:58<06:24, \u001b[1m13.\u001b[0m25s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m44\u001b[0m%|████▍     |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m22\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:12<06:17, \u001b[1m13.\u001b[0m48s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.003621', 'grad_norm': '2.035', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '5.8e-09', 'num_tokens': '1.393e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '30.69', 'completions/min_length': '18', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '30.68', 'completions/min_length': '18', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '45', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '30.69',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '30.68',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '18',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '45',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7109',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4551',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9607',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.1956', 'reward': '1.191',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.4948', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '9.293e-05', 'entropy': '0.01876', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '13.59', 'epoch': '0.176'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7031',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4587',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.945', 'rewards/format_reward/std': \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.2295', 'reward': '1.176', 'reward_std': '0.5124',                \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'frac_reward_zero_std': '0', 'kl': '0.0001385', 'entropy':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.01903', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',   \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',            \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'clip_ratio/region_mean': '0', 'step_time': '13.99', 'epoch':       \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.176'\u001b[1m}\u001b[0m                                                            \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -4990,49 +4979,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:35:42] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             44%|████▍     | 22/50 [05:10<06:14, 13.38s/it]#015 46%|████▌     |                    \n",
-       "                             23/50 [05:26<06:23, 14.21s/it]#015                                                    \n",
-       "                             #015{'loss': '0.001856', 'grad_norm': '1.338', 'learning_rate':                       \n",
+       "
[06/26/26 11:08:07] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             44%|████▍     | 22/50 [05:12<06:17, 13.48s/it]#015 46%|████▌     |                    \n",
+       "                             23/50 [05:28<06:22, 14.15s/it]#015                                                    \n",
+       "                             #015{'loss': '0.00291', 'grad_norm': '2.108', 'learning_rate':                        \n",
        "                             '5.6e-09', 'num_tokens': '1.461e+06', 'completions/mean_length':                      \n",
-       "                             '38.4', 'completions/min_length': '17', 'completions/max_length':                     \n",
+       "                             '38.35', 'completions/min_length': '17', 'completions/max_length':                    \n",
        "                             '128', 'completions/clipped_ratio': '0.0625',                                         \n",
-       "                             'completions/mean_terminated_length': '32.43',                                        \n",
+       "                             'completions/mean_terminated_length': '32.38',                                        \n",
        "                             'completions/min_terminated_length': '17',                                            \n",
        "                             'completions/max_terminated_length': '47',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.8359',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.3718',                                             \n",
-       "                             'rewards/format_reward/mean': '0.882', 'rewards/format_reward/std':                   \n",
-       "                             '0.3252', 'reward': '1.277', 'reward_std': '0.5139',                                  \n",
-       "                             'frac_reward_zero_std': '0', 'kl': '7.992e-06', 'entropy':                            \n",
-       "                             '0.01434', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',                     \n",
-       "                             'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',                              \n",
-       "                             'clip_ratio/region_mean': '0', 'step_time': '16.14', 'epoch':                         \n",
-       "                             '0.184'}                                                                              \n",
+       "                             'rewards/tool_call_reward/mean': '0.8047',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.398',                                              \n",
+       "                             'rewards/format_reward/mean': '0.8349',                                               \n",
+       "                             'rewards/format_reward/std': '0.3741', 'reward': '1.222',                             \n",
+       "                             'reward_std': '0.5721', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.0001697', 'entropy': '0.01636', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '15.71', 'epoch': '0.184'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:35:42]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027683;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027684;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m44\u001b[0m%|████▍     | \u001b[1m22\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:10<06:14, \u001b[1m13.\u001b[0m38s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m46\u001b[0m%|████▌     |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m23\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:26<06:23, \u001b[1m14.\u001b[0m21s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.001856', 'grad_norm': '1.338', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:08:07]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029387;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029388;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m44\u001b[0m%|████▍     | \u001b[1m22\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:12<06:17, \u001b[1m13.\u001b[0m48s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m46\u001b[0m%|████▌     |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m23\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:28<06:22, \u001b[1m14.\u001b[0m15s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.00291', 'grad_norm': '2.108', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '5.6e-09', 'num_tokens': '1.461e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '38.4', 'completions/min_length': '17', 'completions/max_length':   \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '38.35', 'completions/min_length': '17', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '128', 'completions/clipped_ratio': '0.0625',                       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '32.43',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '32.38',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '17',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '47',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8359',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.3718',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.882', 'rewards/format_reward/std': \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.3252', 'reward': '1.277', 'reward_std': '0.5139',                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'frac_reward_zero_std': '0', 'kl': '7.992e-06', 'entropy':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.01434', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/region_mean': '0', 'step_time': '16.14', 'epoch':       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.184'\u001b[1m}\u001b[0m                                                            \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8047',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.398',                            \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8349',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.3741', 'reward': '1.222',           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.5721', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.0001697', 'entropy': '0.01636', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'step_time': '15.71', 'epoch': '0.184'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -5041,49 +5030,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:35:58] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             46%|████▌     | 23/50 [05:26<06:23, 14.21s/it]#015 48%|████▊     |                    \n",
-       "                             24/50 [05:40<06:02, 13.94s/it]#015                                                    \n",
-       "                             #015{'loss': '0.001372', 'grad_norm': '1.178', 'learning_rate':                       \n",
+       "
[06/26/26 11:08:18] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             46%|████▌     | 23/50 [05:28<06:22, 14.15s/it]#015 48%|████▊     |                    \n",
+       "                             24/50 [05:41<06:02, 13.93s/it]#015                                                    \n",
+       "                             #015{'loss': '0.001389', 'grad_norm': '1.385', 'learning_rate':                       \n",
        "                             '5.4e-09', 'num_tokens': '1.526e+06', 'completions/mean_length':                      \n",
-       "                             '39.73', 'completions/min_length': '21', 'completions/max_length':                    \n",
+       "                             '39.6', 'completions/min_length': '21', 'completions/max_length':                     \n",
        "                             '84', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '39.73',                                        \n",
+       "                             'completions/mean_terminated_length': '39.6',                                         \n",
        "                             'completions/min_terminated_length': '21',                                            \n",
        "                             'completions/max_terminated_length': '84',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.5547',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.499',                                              \n",
-       "                             'rewards/format_reward/mean': '0.9214',                                               \n",
-       "                             'rewards/format_reward/std': '0.2711', 'reward': '1.015',                             \n",
-       "                             'reward_std': '0.5579', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.0002224', 'entropy': '0.06501', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'rewards/tool_call_reward/mean': '0.5625',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.498',                                              \n",
+       "                             'rewards/format_reward/mean': '0.9135',                                               \n",
+       "                             'rewards/format_reward/std': '0.2831', 'reward': '1.019',                             \n",
+       "                             'reward_std': '0.5631', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.0001944', 'entropy': '0.06599', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '13.29', 'epoch': '0.192'}                                               \n",
+       "                             'step_time': '13.4', 'epoch': '0.192'}                                                \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:35:58]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027691;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027692;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m46\u001b[0m%|████▌     | \u001b[1m23\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:26<06:23, \u001b[1m14.\u001b[0m21s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m48\u001b[0m%|████▊     |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m24\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:40<06:02, \u001b[1m13.\u001b[0m94s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.001372', 'grad_norm': '1.178', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:08:18]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029395;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029396;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m46\u001b[0m%|████▌     | \u001b[1m23\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:28<06:22, \u001b[1m14.\u001b[0m15s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m48\u001b[0m%|████▊     |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m24\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:41<06:02, \u001b[1m13.\u001b[0m93s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.001389', 'grad_norm': '1.385', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '5.4e-09', 'num_tokens': '1.526e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '39.73', 'completions/min_length': '21', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '39.6', 'completions/min_length': '21', 'completions/max_length':   \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '84', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '39.73',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '39.6',                       \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '21',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '84',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.5547',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.499',                            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9214',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2711', 'reward': '1.015',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5579', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0002224', 'entropy': '0.06501', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.5625',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.498',                            \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9135',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2831', 'reward': '1.019',           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.5631', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.0001944', 'entropy': '0.06599', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '13.29', 'epoch': '0.192'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '13.4', 'epoch': '0.192'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -5092,49 +5081,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:36:14] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             48%|████▊     | 24/50 [05:40<06:02, 13.94s/it]#015 50%|█████     |                    \n",
-       "                             25/50 [05:54<05:53, 14.16s/it]#015                                                    \n",
-       "                             #015{'loss': '0.002559', 'grad_norm': '2.119', 'learning_rate':                       \n",
+       "
[06/26/26 11:08:34] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             48%|████▊     | 24/50 [05:41<06:02, 13.93s/it]#015 50%|█████     |                    \n",
+       "                             25/50 [05:56<05:54, 14.18s/it]#015                                                    \n",
+       "                             #015{'loss': '0.01243', 'grad_norm': '4.469', 'learning_rate':                        \n",
        "                             '5.2e-09', 'num_tokens': '1.586e+06', 'completions/mean_length':                      \n",
-       "                             '36.06', 'completions/min_length': '19', 'completions/max_length':                    \n",
+       "                             '36.21', 'completions/min_length': '19', 'completions/max_length':                    \n",
        "                             '56', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '36.06',                                        \n",
+       "                             'completions/mean_terminated_length': '36.21',                                        \n",
        "                             'completions/min_terminated_length': '19',                                            \n",
        "                             'completions/max_terminated_length': '56',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.8047',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.398',                                              \n",
-       "                             'rewards/format_reward/mean': '0.9136',                                               \n",
-       "                             'rewards/format_reward/std': '0.2829', 'reward': '1.261',                             \n",
-       "                             'reward_std': '0.4985', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '6.807e-05', 'entropy': '0.01488', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'rewards/tool_call_reward/mean': '0.8203',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.3854',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9372',                                               \n",
+       "                             'rewards/format_reward/std': '0.2443', 'reward': '1.289',                             \n",
+       "                             'reward_std': '0.4641', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.0001467', 'entropy': '0.01645', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '14.66', 'epoch': '0.2'}                                                 \n",
+       "                             'step_time': '14.75', 'epoch': '0.2'}                                                 \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:36:14]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027699;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027700;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m48\u001b[0m%|████▊     | \u001b[1m24\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:40<06:02, \u001b[1m13.\u001b[0m94s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m50\u001b[0m%|█████     |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m25\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:54<05:53, \u001b[1m14.\u001b[0m16s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.002559', 'grad_norm': '2.119', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:08:34]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029403;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029404;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m48\u001b[0m%|████▊     | \u001b[1m24\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:41<06:02, \u001b[1m13.\u001b[0m93s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m50\u001b[0m%|█████     |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m25\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:56<05:54, \u001b[1m14.\u001b[0m18s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.01243', 'grad_norm': '4.469', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '5.2e-09', 'num_tokens': '1.586e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '36.06', 'completions/min_length': '19', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '36.21', 'completions/min_length': '19', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '56', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '36.06',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '36.21',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '19',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '56',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8047',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.398',                            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9136',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2829', 'reward': '1.261',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.4985', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '6.807e-05', 'entropy': '0.01488', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8203',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.3854',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9372',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2443', 'reward': '1.289',           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.4641', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.0001467', 'entropy': '0.01645', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '14.66', 'epoch': '0.2'\u001b[1m}\u001b[0m                               \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '14.75', 'epoch': '0.2'\u001b[1m}\u001b[0m                               \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -5143,49 +5132,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:36:25] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             50%|█████     | 25/50 [05:54<05:53, 14.16s/it]#015 52%|█████▏    |                    \n",
-       "                             26/50 [06:08<05:35, 14.00s/it]#015                                                    \n",
-       "                             #015{'loss': '0.003904', 'grad_norm': '3.429', 'learning_rate':                       \n",
+       "
[06/26/26 11:08:50] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             50%|█████     | 25/50 [05:56<05:54, 14.18s/it]#015 52%|█████▏    |                    \n",
+       "                             26/50 [06:12<05:51, 14.63s/it]#015                                                    \n",
+       "                             #015{'loss': '0.01857', 'grad_norm': '4.155', 'learning_rate':                        \n",
        "                             '5e-09', 'num_tokens': '1.65e+06', 'completions/mean_length':                         \n",
-       "                             '32.04', 'completions/min_length': '21', 'completions/max_length':                    \n",
-       "                             '81', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '32.04',                                        \n",
+       "                             '32.46', 'completions/min_length': '21', 'completions/max_length':                    \n",
+       "                             '128', 'completions/clipped_ratio': '0.007812',                                       \n",
+       "                             'completions/mean_terminated_length': '31.71',                                        \n",
        "                             'completions/min_terminated_length': '21',                                            \n",
-       "                             'completions/max_terminated_length': '81',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7109',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4551',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9292',                                               \n",
-       "                             'rewards/format_reward/std': '0.2583', 'reward': '1.176',                             \n",
-       "                             'reward_std': '0.5239', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '7.857e-05', 'entropy': '0.0258', 'clip_ratio/low_mean': '0',                         \n",
+       "                             'completions/max_terminated_length': '66',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7344',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4434',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9369',                                               \n",
+       "                             'rewards/format_reward/std': '0.2452', 'reward': '1.203',                             \n",
+       "                             'reward_std': '0.5082', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.0001655', 'entropy': '0.0276', 'clip_ratio/low_mean': '0',                         \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '13.61', 'epoch': '0.208'}                                               \n",
+       "                             'step_time': '15.67', 'epoch': '0.208'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:36:25]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027707;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027708;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m50\u001b[0m%|█████     | \u001b[1m25\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:54<05:53, \u001b[1m14.\u001b[0m16s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m52\u001b[0m%|█████▏    |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m26\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:08<05:35, \u001b[1m14.\u001b[0m00s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.003904', 'grad_norm': '3.429', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:08:50]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029411;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029412;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m50\u001b[0m%|█████     | \u001b[1m25\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:56<05:54, \u001b[1m14.\u001b[0m18s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m52\u001b[0m%|█████▏    |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m26\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:12<05:51, \u001b[1m14.\u001b[0m63s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.01857', 'grad_norm': '4.155', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '5e-09', 'num_tokens': '1.65e+06', 'completions/mean_length':       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '32.04', 'completions/min_length': '21', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '81', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '32.04',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '32.46', 'completions/min_length': '21', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '128', 'completions/clipped_ratio': '0.007812',                     \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '31.71',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '21',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '81',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7109',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4551',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9292',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2583', 'reward': '1.176',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5239', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '7.857e-05', 'entropy': '0.0258', 'clip_ratio/low_mean': '0',       \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '66',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7344',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4434',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9369',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2452', 'reward': '1.203',           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.5082', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.0001655', 'entropy': '0.0276', 'clip_ratio/low_mean': '0',       \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '13.61', 'epoch': '0.208'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '15.67', 'epoch': '0.208'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -5194,49 +5183,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:36:41] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             52%|█████▏    | 26/50 [06:08<05:35, 14.00s/it]#015 54%|█████▍    |                    \n",
-       "                             27/50 [06:24<05:32, 14.46s/it]#015                                                    \n",
-       "                             #015{'loss': '0.04762', 'grad_norm': '1.92', 'learning_rate':                         \n",
+       "
[06/26/26 11:09:07] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             52%|█████▏    | 26/50 [06:12<05:51, 14.63s/it]#015 54%|█████▍    |                    \n",
+       "                             27/50 [06:27<05:43, 14.92s/it]#015                                                    \n",
+       "                             #015{'loss': '0.03352', 'grad_norm': '1.645', 'learning_rate':                        \n",
        "                             '4.8e-09', 'num_tokens': '1.719e+06', 'completions/mean_length':                      \n",
-       "                             '38.88', 'completions/min_length': '20', 'completions/max_length':                    \n",
-       "                             '128', 'completions/clipped_ratio': '0.04688',                                        \n",
-       "                             'completions/mean_terminated_length': '34.49',                                        \n",
+       "                             '39.02', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '128', 'completions/clipped_ratio': '0.05469',                                        \n",
+       "                             'completions/mean_terminated_length': '33.88',                                        \n",
        "                             'completions/min_terminated_length': '20',                                            \n",
        "                             'completions/max_terminated_length': '59',                                            \n",
        "                             'rewards/tool_call_reward/mean': '0.6953',                                            \n",
        "                             'rewards/tool_call_reward/std': '0.4621',                                             \n",
-       "                             'rewards/format_reward/mean': '0.8661',                                               \n",
-       "                             'rewards/format_reward/std': '0.3434', 'reward': '1.128',                             \n",
-       "                             'reward_std': '0.5804', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '5.935e-05', 'entropy': '0.01181', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '15.51', 'epoch': '0.216'}                                               \n",
+       "                             'rewards/format_reward/mean': '0.874', 'rewards/format_reward/std':                   \n",
+       "                             '0.3347', 'reward': '1.132', 'reward_std': '0.5743',                                  \n",
+       "                             'frac_reward_zero_std': '0', 'kl': '2.56e-05', 'entropy':                             \n",
+       "                             '0.01199', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',                     \n",
+       "                             'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',                              \n",
+       "                             'clip_ratio/region_mean': '0', 'step_time': '15.6', 'epoch':                          \n",
+       "                             '0.216'}                                                                              \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:36:41]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027715;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027716;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m52\u001b[0m%|█████▏    | \u001b[1m26\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:08<05:35, \u001b[1m14.\u001b[0m00s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m54\u001b[0m%|█████▍    |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m27\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:24<05:32, \u001b[1m14.\u001b[0m46s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.04762', 'grad_norm': '1.92', 'learning_rate':       \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:09:07]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029419;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029420;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m52\u001b[0m%|█████▏    | \u001b[1m26\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:12<05:51, \u001b[1m14.\u001b[0m63s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m54\u001b[0m%|█████▍    |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m27\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:27<05:43, \u001b[1m14.\u001b[0m92s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.03352', 'grad_norm': '1.645', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '4.8e-09', 'num_tokens': '1.719e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '38.88', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '128', 'completions/clipped_ratio': '0.04688',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '34.49',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '39.02', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '128', 'completions/clipped_ratio': '0.05469',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '33.88',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '59',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.6953',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4621',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8661',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.3434', 'reward': '1.128',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5804', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '5.935e-05', 'entropy': '0.01181', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '15.51', 'epoch': '0.216'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.874', 'rewards/format_reward/std': \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.3347', 'reward': '1.132', 'reward_std': '0.5743',                \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'frac_reward_zero_std': '0', 'kl': '2.56e-05', 'entropy':           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.01199', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',   \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',            \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'clip_ratio/region_mean': '0', 'step_time': '15.6', 'epoch':        \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.216'\u001b[1m}\u001b[0m                                                            \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -5245,49 +5234,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:36:52] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             54%|█████▍    | 27/50 [06:24<05:32, 14.46s/it]#015 56%|█████▌    |                    \n",
-       "                             28/50 [06:38<05:15, 14.35s/it]#015                                                    \n",
-       "                             #015{'loss': '-0.004456', 'grad_norm': '1.489', 'learning_rate':                      \n",
+       "
[06/26/26 11:09:18] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             54%|█████▍    | 27/50 [06:27<05:43, 14.92s/it]#015 56%|█████▌    |                    \n",
+       "                             28/50 [06:40<05:13, 14.26s/it]#015                                                    \n",
+       "                             #015{'loss': '-0.01125', 'grad_norm': '2.16', 'learning_rate':                        \n",
        "                             '4.6e-09', 'num_tokens': '1.786e+06', 'completions/mean_length':                      \n",
-       "                             '33.55', 'completions/min_length': '19', 'completions/max_length':                    \n",
+       "                             '33.49', 'completions/min_length': '19', 'completions/max_length':                    \n",
        "                             '59', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '33.55',                                        \n",
+       "                             'completions/mean_terminated_length': '33.49',                                        \n",
        "                             'completions/min_terminated_length': '19',                                            \n",
        "                             'completions/max_terminated_length': '59',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.6875',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4653',                                             \n",
-       "                             'rewards/format_reward/mean': '0.8979',                                               \n",
-       "                             'rewards/format_reward/std': '0.3049', 'reward': '1.136',                             \n",
-       "                             'reward_std': '0.5572', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.000104', 'entropy': '0.01498', 'clip_ratio/low_mean': '0',                         \n",
+       "                             'rewards/tool_call_reward/mean': '0.6797',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4684',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9214',                                               \n",
+       "                             'rewards/format_reward/std': '0.2709', 'reward': '1.14',                              \n",
+       "                             'reward_std': '0.54', 'frac_reward_zero_std': '0', 'kl':                              \n",
+       "                             '0.0001186', 'entropy': '0.01506', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '14.11', 'epoch': '0.224'}                                               \n",
+       "                             'step_time': '12.69', 'epoch': '0.224'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:36:52]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027723;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027724;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m54\u001b[0m%|█████▍    | \u001b[1m27\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:24<05:32, \u001b[1m14.\u001b[0m46s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m56\u001b[0m%|█████▌    |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m28\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:38<05:15, \u001b[1m14.\u001b[0m35s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.004456', 'grad_norm': '1.489', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:09:18]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029427;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029428;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m54\u001b[0m%|█████▍    | \u001b[1m27\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:27<05:43, \u001b[1m14.\u001b[0m92s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m56\u001b[0m%|█████▌    |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m28\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:40<05:13, \u001b[1m14.\u001b[0m26s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.01125', 'grad_norm': '2.16', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '4.6e-09', 'num_tokens': '1.786e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '33.55', 'completions/min_length': '19', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '33.49', 'completions/min_length': '19', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '59', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '33.55',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '33.49',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '19',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '59',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.6875',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4653',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8979',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.3049', 'reward': '1.136',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5572', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.000104', 'entropy': '0.01498', 'clip_ratio/low_mean': '0',       \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.6797',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4684',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9214',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2709', 'reward': '1.14',            \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.54', 'frac_reward_zero_std': '0', 'kl':            \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.0001186', 'entropy': '0.01506', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '14.11', 'epoch': '0.224'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '12.69', 'epoch': '0.224'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -5296,49 +5285,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:37:08] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             56%|█████▌    | 28/50 [06:38<05:15, 14.35s/it]#015 58%|█████▊    |                    \n",
-       "                             29/50 [06:50<04:50, 13.85s/it]#015                                                    \n",
-       "                             #015{'loss': '0.002348', 'grad_norm': '2.229', 'learning_rate':                       \n",
+       "
[06/26/26 11:09:29] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             56%|█████▌    | 28/50 [06:40<05:13, 14.26s/it]#015 58%|█████▊    |                    \n",
+       "                             29/50 [06:53<04:50, 13.82s/it]#015                                                    \n",
+       "                             #015{'loss': '0.00335', 'grad_norm': '1.61', 'learning_rate':                         \n",
        "                             '4.4e-09', 'num_tokens': '1.85e+06', 'completions/mean_length':                       \n",
-       "                             '29.97', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '29.99', 'completions/min_length': '20', 'completions/max_length':                    \n",
        "                             '46', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '29.97',                                        \n",
+       "                             'completions/mean_terminated_length': '29.99',                                        \n",
        "                             'completions/min_terminated_length': '20',                                            \n",
        "                             'completions/max_terminated_length': '46',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7891',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4096',                                             \n",
-       "                             'rewards/format_reward/mean': '0.945', 'rewards/format_reward/std':                   \n",
-       "                             '0.2294', 'reward': '1.262', 'reward_std': '0.4739',                                  \n",
-       "                             'frac_reward_zero_std': '0', 'kl': '0.000111', 'entropy':                             \n",
-       "                             '0.03362', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',                     \n",
-       "                             'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',                              \n",
-       "                             'clip_ratio/region_mean': '0', 'step_time': '12.65', 'epoch':                         \n",
-       "                             '0.232'}                                                                              \n",
+       "                             'rewards/tool_call_reward/mean': '0.7969',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4039',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9372',                                               \n",
+       "                             'rewards/format_reward/std': '0.2442', 'reward': '1.265',                             \n",
+       "                             'reward_std': '0.478', 'frac_reward_zero_std': '0', 'kl':                             \n",
+       "                             '0.0001361', 'entropy': '0.03338', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '12.79', 'epoch': '0.232'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:37:08]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027731;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027732;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m56\u001b[0m%|█████▌    | \u001b[1m28\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:38<05:15, \u001b[1m14.\u001b[0m35s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m58\u001b[0m%|█████▊    |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m29\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:50<04:50, \u001b[1m13.\u001b[0m85s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.002348', 'grad_norm': '2.229', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:09:29]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029435;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029436;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m56\u001b[0m%|█████▌    | \u001b[1m28\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:40<05:13, \u001b[1m14.\u001b[0m26s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m58\u001b[0m%|█████▊    |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m29\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:53<04:50, \u001b[1m13.\u001b[0m82s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.00335', 'grad_norm': '1.61', 'learning_rate':       \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '4.4e-09', 'num_tokens': '1.85e+06', 'completions/mean_length':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '29.97', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '29.99', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '46', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '29.97',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '29.99',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '46',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7891',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4096',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.945', 'rewards/format_reward/std': \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.2294', 'reward': '1.262', 'reward_std': '0.4739',                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'frac_reward_zero_std': '0', 'kl': '0.000111', 'entropy':           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.03362', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/region_mean': '0', 'step_time': '12.65', 'epoch':       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.232'\u001b[1m}\u001b[0m                                                            \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7969',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4039',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9372',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2442', 'reward': '1.265',           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.478', 'frac_reward_zero_std': '0', 'kl':           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.0001361', 'entropy': '0.03338', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'step_time': '12.79', 'epoch': '0.232'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -5347,49 +5336,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:37:24] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             58%|█████▊    | 29/50 [06:50<04:50, 13.85s/it]#015 60%|██████    |                    \n",
-       "                             30/50 [07:06<04:47, 14.38s/it]#015                                                    \n",
-       "                             #015{'loss': '0.003175', 'grad_norm': '0.7581', 'learning_rate':                      \n",
+       "
[06/26/26 11:09:45] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             58%|█████▊    | 29/50 [06:53<04:50, 13.82s/it]#015 60%|██████    |                    \n",
+       "                             30/50 [07:09<04:49, 14.46s/it]#015                                                    \n",
+       "                             #015{'loss': '0.003962', 'grad_norm': '1.33', 'learning_rate':                        \n",
        "                             '4.2e-09', 'num_tokens': '1.919e+06', 'completions/mean_length':                      \n",
-       "                             '41.95', 'completions/min_length': '19', 'completions/max_length':                    \n",
+       "                             '42.09', 'completions/min_length': '19', 'completions/max_length':                    \n",
        "                             '128', 'completions/clipped_ratio': '0.0625',                                         \n",
-       "                             'completions/mean_terminated_length': '36.21',                                        \n",
+       "                             'completions/mean_terminated_length': '36.36',                                        \n",
        "                             'completions/min_terminated_length': '19',                                            \n",
        "                             'completions/max_terminated_length': '90',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.6953',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4621',                                             \n",
-       "                             'rewards/format_reward/mean': '0.8732',                                               \n",
-       "                             'rewards/format_reward/std': '0.3369', 'reward': '1.132',                             \n",
-       "                             'reward_std': '0.5751', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.0001467', 'entropy': '0.03984', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'rewards/tool_call_reward/mean': '0.6641',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4742',                                             \n",
+       "                             'rewards/format_reward/mean': '0.8181',                                               \n",
+       "                             'rewards/format_reward/std': '0.3903', 'reward': '1.073',                             \n",
+       "                             'reward_std': '0.6202', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.0001873', 'entropy': '0.04145', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '15.6', 'epoch': '0.24'}                                                 \n",
+       "                             'step_time': '15.93', 'epoch': '0.24'}                                                \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:37:24]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027739;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027740;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m58\u001b[0m%|█████▊    | \u001b[1m29\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:50<04:50, \u001b[1m13.\u001b[0m85s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m60\u001b[0m%|██████    |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m30\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:06<04:47, \u001b[1m14.\u001b[0m38s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.003175', 'grad_norm': '0.7581', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:09:45]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029443;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029444;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m58\u001b[0m%|█████▊    | \u001b[1m29\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:53<04:50, \u001b[1m13.\u001b[0m82s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m60\u001b[0m%|██████    |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m30\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:09<04:49, \u001b[1m14.\u001b[0m46s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.003962', 'grad_norm': '1.33', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '4.2e-09', 'num_tokens': '1.919e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '41.95', 'completions/min_length': '19', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '42.09', 'completions/min_length': '19', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '128', 'completions/clipped_ratio': '0.0625',                       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '36.21',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '36.36',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '19',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '90',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.6953',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4621',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8732',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.3369', 'reward': '1.132',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5751', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0001467', 'entropy': '0.03984', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.6641',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4742',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8181',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.3903', 'reward': '1.073',           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.6202', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.0001873', 'entropy': '0.04145', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '15.6', 'epoch': '0.24'\u001b[1m}\u001b[0m                               \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '15.93', 'epoch': '0.24'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -5398,49 +5387,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:37:35] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             60%|██████    | 30/50 [07:06<04:47, 14.38s/it]#015 62%|██████▏   |                    \n",
-       "                             31/50 [07:18<04:20, 13.73s/it]#015                                                    \n",
-       "                             #015{'loss': '-0.005809', 'grad_norm': '2.727', 'learning_rate':                      \n",
+       "
[06/26/26 11:09:55] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             60%|██████    | 30/50 [07:09<04:49, 14.46s/it]#015 62%|██████▏   |                    \n",
+       "                             31/50 [07:20<04:17, 13.56s/it]#015                                                    \n",
+       "                             #015{'loss': '-0.005677', 'grad_norm': '2.497', 'learning_rate':                      \n",
        "                             '4e-09', 'num_tokens': '1.974e+06', 'completions/mean_length':                        \n",
        "                             '32.5', 'completions/min_length': '21', 'completions/max_length':                     \n",
        "                             '41', 'completions/clipped_ratio': '0',                                               \n",
        "                             'completions/mean_terminated_length': '32.5',                                         \n",
        "                             'completions/min_terminated_length': '21',                                            \n",
        "                             'completions/max_terminated_length': '41',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7734',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4203',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9214',                                               \n",
-       "                             'rewards/format_reward/std': '0.2709', 'reward': '1.234',                             \n",
-       "                             'reward_std': '0.5062', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '8.09e-05', 'entropy': '0.01413', 'clip_ratio/low_mean': '0',                         \n",
+       "                             'rewards/tool_call_reward/mean': '0.7656',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4253',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9136',                                               \n",
+       "                             'rewards/format_reward/std': '0.2829', 'reward': '1.222',                             \n",
+       "                             'reward_std': '0.5172', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.0001616', 'entropy': '0.01579', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '12.19', 'epoch': '0.248'}                                               \n",
+       "                             'step_time': '11.46', 'epoch': '0.248'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:37:35]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027747;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027748;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m60\u001b[0m%|██████    | \u001b[1m30\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:06<04:47, \u001b[1m14.\u001b[0m38s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m62\u001b[0m%|██████▏   |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m31\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:18<04:20, \u001b[1m13.\u001b[0m73s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.005809', 'grad_norm': '2.727', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:09:55]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029451;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029452;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m60\u001b[0m%|██████    | \u001b[1m30\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:09<04:49, \u001b[1m14.\u001b[0m46s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m62\u001b[0m%|██████▏   |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m31\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:20<04:17, \u001b[1m13.\u001b[0m56s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.005677', 'grad_norm': '2.497', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '4e-09', 'num_tokens': '1.974e+06', 'completions/mean_length':      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '32.5', 'completions/min_length': '21', 'completions/max_length':   \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '41', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '32.5',                       \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '21',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '41',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7734',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4203',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9214',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2709', 'reward': '1.234',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5062', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '8.09e-05', 'entropy': '0.01413', 'clip_ratio/low_mean': '0',       \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7656',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4253',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9136',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2829', 'reward': '1.222',           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.5172', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.0001616', 'entropy': '0.01579', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '12.19', 'epoch': '0.248'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '11.46', 'epoch': '0.248'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -5449,49 +5438,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:37:51] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             62%|██████▏   | 31/50 [07:18<04:20, 13.73s/it]#015 64%|██████▍   |                    \n",
-       "                             32/50 [07:32<04:10, 13.89s/it]#015                                                    \n",
-       "                             #015{'loss': '0.0004126', 'grad_norm': '2.2', 'learning_rate':                        \n",
+       "
[06/26/26 11:10:11] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             62%|██████▏   | 31/50 [07:20<04:17, 13.56s/it]#015 64%|██████▍   |                    \n",
+       "                             32/50 [07:36<04:13, 14.07s/it]#015                                                    \n",
+       "                             #015{'loss': '-0.0004137', 'grad_norm': '2.191', 'learning_rate':                     \n",
        "                             '3.8e-09', 'num_tokens': '2.054e+06', 'completions/mean_length':                      \n",
-       "                             '35.3', 'completions/min_length': '22', 'completions/max_length':                     \n",
-       "                             '62', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '35.3',                                         \n",
+       "                             '35.31', 'completions/min_length': '22', 'completions/max_length':                    \n",
+       "                             '61', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '35.31',                                        \n",
        "                             'completions/min_terminated_length': '22',                                            \n",
-       "                             'completions/max_terminated_length': '62',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7812',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.415',                                              \n",
-       "                             'rewards/format_reward/mean': '0.9371',                                               \n",
-       "                             'rewards/format_reward/std': '0.2446', 'reward': '1.25',                              \n",
-       "                             'reward_std': '0.4866', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '8.267e-05', 'entropy': '0.022', 'clip_ratio/low_mean': '0',                          \n",
+       "                             'completions/max_terminated_length': '61',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7891',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4096',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9528',                                               \n",
+       "                             'rewards/format_reward/std': '0.2136', 'reward': '1.265',                             \n",
+       "                             'reward_std': '0.4655', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '4.369e-05', 'entropy': '0.0221', 'clip_ratio/low_mean': '0',                         \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '14.27', 'epoch': '0.256'}                                               \n",
+       "                             'step_time': '15.24', 'epoch': '0.256'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:37:51]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027755;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027756;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m62\u001b[0m%|██████▏   | \u001b[1m31\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:18<04:20, \u001b[1m13.\u001b[0m73s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m64\u001b[0m%|██████▍   |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m32\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:32<04:10, \u001b[1m13.\u001b[0m89s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.0004126', 'grad_norm': '2.2', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:10:11]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029459;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029460;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m62\u001b[0m%|██████▏   | \u001b[1m31\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:20<04:17, \u001b[1m13.\u001b[0m56s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m64\u001b[0m%|██████▍   |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m32\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:36<04:13, \u001b[1m14.\u001b[0m07s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.0004137', 'grad_norm': '2.191', 'learning_rate':   \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '3.8e-09', 'num_tokens': '2.054e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '35.3', 'completions/min_length': '22', 'completions/max_length':   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '62', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '35.3',                       \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '35.31', 'completions/min_length': '22', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '61', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '35.31',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '22',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '62',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7812',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.415',                            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9371',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2446', 'reward': '1.25',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.4866', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '8.267e-05', 'entropy': '0.022', 'clip_ratio/low_mean': '0',        \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '61',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7891',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4096',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9528',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2136', 'reward': '1.265',           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.4655', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '4.369e-05', 'entropy': '0.0221', 'clip_ratio/low_mean': '0',       \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '14.27', 'epoch': '0.256'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '15.24', 'epoch': '0.256'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -5500,49 +5489,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:38:01] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             64%|██████▍   | 32/50 [07:32<04:10, 13.89s/it]#015 66%|██████▌   |                    \n",
-       "                             33/50 [07:46<03:55, 13.87s/it]#015                                                    \n",
-       "                             #015{'loss': '0.01859', 'grad_norm': '2.834', 'learning_rate':                        \n",
+       "
[06/26/26 11:10:27] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             64%|██████▍   | 32/50 [07:36<04:13, 14.07s/it]#015 66%|██████▌   |                    \n",
+       "                             33/50 [07:49<03:56, 13.92s/it]#015                                                    \n",
+       "                             #015{'loss': '0.007093', 'grad_norm': '2.214', 'learning_rate':                       \n",
        "                             '3.6e-09', 'num_tokens': '2.112e+06', 'completions/mean_length':                      \n",
-       "                             '38.46', 'completions/min_length': '20', 'completions/max_length':                    \n",
-       "                             '94', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '38.46',                                        \n",
+       "                             '37.85', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '84', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '37.85',                                        \n",
        "                             'completions/min_terminated_length': '20',                                            \n",
-       "                             'completions/max_terminated_length': '94',                                            \n",
+       "                             'completions/max_terminated_length': '84',                                            \n",
        "                             'rewards/tool_call_reward/mean': '0.7734',                                            \n",
        "                             'rewards/tool_call_reward/std': '0.4203',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9229',                                               \n",
-       "                             'rewards/format_reward/std': '0.2665', 'reward': '1.235',                             \n",
-       "                             'reward_std': '0.5045', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.0002194', 'entropy': '0.0316', 'clip_ratio/low_mean': '0',                         \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '13.8', 'epoch': '0.264'}                                                \n",
+       "                             'rewards/format_reward/mean': '0.945', 'rewards/format_reward/std':                   \n",
+       "                             '0.2294', 'reward': '1.246', 'reward_std': '0.4823',                                  \n",
+       "                             'frac_reward_zero_std': '0', 'kl': '0.0002727', 'entropy':                            \n",
+       "                             '0.02825', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',                     \n",
+       "                             'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',                              \n",
+       "                             'clip_ratio/region_mean': '0', 'step_time': '13.56', 'epoch':                         \n",
+       "                             '0.264'}                                                                              \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:38:01]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027763;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027764;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m64\u001b[0m%|██████▍   | \u001b[1m32\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:32<04:10, \u001b[1m13.\u001b[0m89s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m66\u001b[0m%|██████▌   |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m33\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:46<03:55, \u001b[1m13.\u001b[0m87s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.01859', 'grad_norm': '2.834', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:10:27]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029467;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029468;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m64\u001b[0m%|██████▍   | \u001b[1m32\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:36<04:13, \u001b[1m14.\u001b[0m07s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m66\u001b[0m%|██████▌   |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m33\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:49<03:56, \u001b[1m13.\u001b[0m92s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.007093', 'grad_norm': '2.214', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '3.6e-09', 'num_tokens': '2.112e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '38.46', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '94', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '38.46',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '37.85', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '84', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '37.85',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '94',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '84',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7734',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4203',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9229',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2665', 'reward': '1.235',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5045', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0002194', 'entropy': '0.0316', 'clip_ratio/low_mean': '0',       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '13.8', 'epoch': '0.264'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.945', 'rewards/format_reward/std': \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.2294', 'reward': '1.246', 'reward_std': '0.4823',                \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'frac_reward_zero_std': '0', 'kl': '0.0002727', 'entropy':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.02825', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',   \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',            \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'clip_ratio/region_mean': '0', 'step_time': '13.56', 'epoch':       \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.264'\u001b[1m}\u001b[0m                                                            \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -5551,49 +5540,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:38:12] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             66%|██████▌   | 33/50 [07:46<03:55, 13.87s/it]#015 68%|██████▊   |                    \n",
-       "                             34/50 [07:59<03:35, 13.50s/it]#015                                                    \n",
-       "                             #015{'loss': '0.02429', 'grad_norm': '3.956', 'learning_rate':                        \n",
+       "
[06/26/26 11:10:39] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             66%|██████▌   | 33/50 [07:49<03:56, 13.92s/it]#015 68%|██████▊   |                    \n",
+       "                             34/50 [08:02<03:35, 13.44s/it]#015                                                    \n",
+       "                             #015{'loss': '0.01603', 'grad_norm': '4.95', 'learning_rate':                         \n",
        "                             '3.4e-09', 'num_tokens': '2.19e+06', 'completions/mean_length':                       \n",
-       "                             '29.49', 'completions/min_length': '20', 'completions/max_length':                    \n",
-       "                             '60', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '29.49',                                        \n",
+       "                             '29.18', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '50', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '29.18',                                        \n",
        "                             'completions/min_terminated_length': '20',                                            \n",
-       "                             'completions/max_terminated_length': '60',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7734',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4203',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9293',                                               \n",
-       "                             'rewards/format_reward/std': '0.2579', 'reward': '1.238',                             \n",
-       "                             'reward_std': '0.4983', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '3.041e-05', 'entropy': '0.022', 'clip_ratio/low_mean': '0',                          \n",
+       "                             'completions/max_terminated_length': '50',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.75',                                              \n",
+       "                             'rewards/tool_call_reward/std': '0.4347',                                             \n",
+       "                             'rewards/format_reward/mean': '0.8979',                                               \n",
+       "                             'rewards/format_reward/std': '0.3048', 'reward': '1.199',                             \n",
+       "                             'reward_std': '0.5379', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.000156', 'entropy': '0.0227', 'clip_ratio/low_mean': '0',                          \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '12.63', 'epoch': '0.272'}                                               \n",
+       "                             'step_time': '12.3', 'epoch': '0.272'}                                                \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:38:12]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027771;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027772;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m66\u001b[0m%|██████▌   | \u001b[1m33\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:46<03:55, \u001b[1m13.\u001b[0m87s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m68\u001b[0m%|██████▊   |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m34\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:59<03:35, \u001b[1m13.\u001b[0m50s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.02429', 'grad_norm': '3.956', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:10:39]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029475;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029476;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m66\u001b[0m%|██████▌   | \u001b[1m33\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:49<03:56, \u001b[1m13.\u001b[0m92s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m68\u001b[0m%|██████▊   |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m34\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m08:02<03:35, \u001b[1m13.\u001b[0m44s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.01603', 'grad_norm': '4.95', 'learning_rate':       \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '3.4e-09', 'num_tokens': '2.19e+06', 'completions/mean_length':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '29.49', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '60', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '29.49',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '29.18', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '50', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '29.18',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '60',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7734',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4203',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9293',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2579', 'reward': '1.238',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.4983', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '3.041e-05', 'entropy': '0.022', 'clip_ratio/low_mean': '0',        \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '50',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.75',                            \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4347',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8979',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.3048', 'reward': '1.199',           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.5379', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.000156', 'entropy': '0.0227', 'clip_ratio/low_mean': '0',        \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '12.63', 'epoch': '0.272'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '12.3', 'epoch': '0.272'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -5602,49 +5591,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:38:29] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             68%|██████▊   | 34/50 [07:59<03:35, 13.50s/it]#015 70%|███████   |                    \n",
-       "                             35/50 [08:14<03:27, 13.83s/it]#015                                                    \n",
-       "                             #015{'loss': '0.002602', 'grad_norm': '2.153', 'learning_rate':                       \n",
+       "
[06/26/26 11:10:55] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             68%|██████▊   | 34/50 [08:02<03:35, 13.44s/it]#015 70%|███████   |                    \n",
+       "                             35/50 [08:17<03:29, 13.96s/it]#015                                                    \n",
+       "                             #015{'loss': '0.002534', 'grad_norm': '1.978', 'learning_rate':                       \n",
        "                             '3.2e-09', 'num_tokens': '2.246e+06', 'completions/mean_length':                      \n",
        "                             '30.71', 'completions/min_length': '20', 'completions/max_length':                    \n",
        "                             '57', 'completions/clipped_ratio': '0',                                               \n",
        "                             'completions/mean_terminated_length': '30.71',                                        \n",
        "                             'completions/min_terminated_length': '20',                                            \n",
        "                             'completions/max_terminated_length': '57',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.8438',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.3645',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9136',                                               \n",
-       "                             'rewards/format_reward/std': '0.283', 'reward': '1.301',                              \n",
-       "                             'reward_std': '0.4758', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '1.009e-05', 'entropy': '0.0104', 'clip_ratio/low_mean': '0',                         \n",
+       "                             'rewards/tool_call_reward/mean': '0.8359',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.3718',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9058',                                               \n",
+       "                             'rewards/format_reward/std': '0.2941', 'reward': '1.289',                             \n",
+       "                             'reward_std': '0.4891', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '5.05e-05', 'entropy': '0.01136', 'clip_ratio/low_mean': '0',                         \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '14.59', 'epoch': '0.28'}                                                \n",
+       "                             'step_time': '15.18', 'epoch': '0.28'}                                                \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:38:29]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027779;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027780;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m68\u001b[0m%|██████▊   | \u001b[1m34\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:59<03:35, \u001b[1m13.\u001b[0m50s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m70\u001b[0m%|███████   |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m35\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m08:14<03:27, \u001b[1m13.\u001b[0m83s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.002602', 'grad_norm': '2.153', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:10:55]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029483;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029484;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m68\u001b[0m%|██████▊   | \u001b[1m34\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m08:02<03:35, \u001b[1m13.\u001b[0m44s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m70\u001b[0m%|███████   |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m35\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m08:17<03:29, \u001b[1m13.\u001b[0m96s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.002534', 'grad_norm': '1.978', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '3.2e-09', 'num_tokens': '2.246e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '30.71', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '57', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '30.71',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '57',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8438',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.3645',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9136',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.283', 'reward': '1.301',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.4758', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '1.009e-05', 'entropy': '0.0104', 'clip_ratio/low_mean': '0',       \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8359',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.3718',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9058',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2941', 'reward': '1.289',           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.4891', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '5.05e-05', 'entropy': '0.01136', 'clip_ratio/low_mean': '0',       \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '14.59', 'epoch': '0.28'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '15.18', 'epoch': '0.28'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -5653,49 +5642,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:38:45] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             70%|███████   | 35/50 [08:14<03:27, 13.83s/it]#015 72%|███████▏  |                    \n",
-       "                             36/50 [08:29<03:18, 14.19s/it]#015                                                    \n",
-       "                             #015{'loss': '0.005399', 'grad_norm': '1.961', 'learning_rate':                       \n",
+       "
[06/26/26 11:11:06] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             70%|███████   | 35/50 [08:17<03:29, 13.96s/it]#015 72%|███████▏  |                    \n",
+       "                             36/50 [08:32<03:20, 14.32s/it]#015                                                    \n",
+       "                             #015{'loss': '0.00398', 'grad_norm': '2.012', 'learning_rate':                        \n",
        "                             '3e-09', 'num_tokens': '2.305e+06', 'completions/mean_length':                        \n",
-       "                             '35.13', 'completions/min_length': '19', 'completions/max_length':                    \n",
+       "                             '34.98', 'completions/min_length': '19', 'completions/max_length':                    \n",
        "                             '119', 'completions/clipped_ratio': '0',                                              \n",
-       "                             'completions/mean_terminated_length': '35.13',                                        \n",
+       "                             'completions/mean_terminated_length': '34.98',                                        \n",
        "                             'completions/min_terminated_length': '19',                                            \n",
        "                             'completions/max_terminated_length': '119',                                           \n",
        "                             'rewards/tool_call_reward/mean': '0.6719',                                            \n",
        "                             'rewards/tool_call_reward/std': '0.4714',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9215',                                               \n",
-       "                             'rewards/format_reward/std': '0.2707', 'reward': '1.133',                             \n",
-       "                             'reward_std': '0.5419', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.0001331', 'entropy': '0.01822', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'rewards/format_reward/mean': '0.9214',                                               \n",
+       "                             'rewards/format_reward/std': '0.2712', 'reward': '1.133',                             \n",
+       "                             'reward_std': '0.5421', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.0001751', 'entropy': '0.01811', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '15', 'epoch': '0.288'}                                                  \n",
+       "                             'step_time': '15.15', 'epoch': '0.288'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:38:45]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027787;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027788;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m70\u001b[0m%|███████   | \u001b[1m35\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m08:14<03:27, \u001b[1m13.\u001b[0m83s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m72\u001b[0m%|███████▏  |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m36\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m08:29<03:18, \u001b[1m14.\u001b[0m19s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.005399', 'grad_norm': '1.961', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:11:06]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029491;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029492;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m70\u001b[0m%|███████   | \u001b[1m35\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m08:17<03:29, \u001b[1m13.\u001b[0m96s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m72\u001b[0m%|███████▏  |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m36\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m08:32<03:20, \u001b[1m14.\u001b[0m32s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.00398', 'grad_norm': '2.012', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '3e-09', 'num_tokens': '2.305e+06', 'completions/mean_length':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '35.13', 'completions/min_length': '19', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '34.98', 'completions/min_length': '19', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '119', 'completions/clipped_ratio': '0',                            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '35.13',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '34.98',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '19',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '119',                         \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.6719',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4714',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9215',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2707', 'reward': '1.133',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5419', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0001331', 'entropy': '0.01822', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9214',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2712', 'reward': '1.133',           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.5421', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.0001751', 'entropy': '0.01811', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '15', 'epoch': '0.288'\u001b[1m}\u001b[0m                                \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '15.15', 'epoch': '0.288'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -5704,49 +5693,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:39:01] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             72%|███████▏  | 36/50 [08:29<03:18, 14.19s/it]#015 74%|███████▍  |                    \n",
-       "                             37/50 [08:44<03:08, 14.47s/it]#015                                                    \n",
-       "                             #015{'loss': '0.02714', 'grad_norm': '4.003', 'learning_rate':                        \n",
+       "
[06/26/26 11:11:23] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             72%|███████▏  | 36/50 [08:32<03:20, 14.32s/it]#015 74%|███████▍  |                    \n",
+       "                             37/50 [08:46<03:04, 14.16s/it]#015                                                    \n",
+       "                             #015{'loss': '0.005717', 'grad_norm': '2.737', 'learning_rate':                       \n",
        "                             '2.8e-09', 'num_tokens': '2.393e+06', 'completions/mean_length':                      \n",
-       "                             '32.17', 'completions/min_length': '22', 'completions/max_length':                    \n",
-       "                             '71', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '32.17',                                        \n",
+       "                             '31.75', 'completions/min_length': '22', 'completions/max_length':                    \n",
+       "                             '48', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '31.75',                                        \n",
        "                             'completions/min_terminated_length': '22',                                            \n",
-       "                             'completions/max_terminated_length': '71',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7656',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4253',                                             \n",
+       "                             'completions/max_terminated_length': '48',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7734',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4203',                                             \n",
        "                             'rewards/format_reward/mean': '0.9214',                                               \n",
-       "                             'rewards/format_reward/std': '0.2712', 'reward': '1.226',                             \n",
-       "                             'reward_std': '0.5098', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.0001479', 'entropy': '0.0256', 'clip_ratio/low_mean': '0',                         \n",
+       "                             'rewards/format_reward/std': '0.2709', 'reward': '1.234',                             \n",
+       "                             'reward_std': '0.5062', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.0001182', 'entropy': '0.02041', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '15.12', 'epoch': '0.296'}                                               \n",
+       "                             'step_time': '13.78', 'epoch': '0.296'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:39:01]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027795;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027796;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m72\u001b[0m%|███████▏  | \u001b[1m36\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m08:29<03:18, \u001b[1m14.\u001b[0m19s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m74\u001b[0m%|███████▍  |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m37\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m08:44<03:08, \u001b[1m14.\u001b[0m47s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.02714', 'grad_norm': '4.003', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:11:23]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029499;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029500;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m72\u001b[0m%|███████▏  | \u001b[1m36\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m08:32<03:20, \u001b[1m14.\u001b[0m32s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m74\u001b[0m%|███████▍  |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m37\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m08:46<03:04, \u001b[1m14.\u001b[0m16s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.005717', 'grad_norm': '2.737', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '2.8e-09', 'num_tokens': '2.393e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '32.17', 'completions/min_length': '22', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '71', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '32.17',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '31.75', 'completions/min_length': '22', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '48', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '31.75',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '22',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '71',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7656',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4253',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '48',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7734',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4203',                           \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9214',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2712', 'reward': '1.226',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5098', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0001479', 'entropy': '0.0256', 'clip_ratio/low_mean': '0',       \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2709', 'reward': '1.234',           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.5062', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.0001182', 'entropy': '0.02041', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '15.12', 'epoch': '0.296'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '13.78', 'epoch': '0.296'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -5755,49 +5744,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:39:17] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             74%|███████▍  | 37/50 [08:44<03:08, 14.47s/it]#015 76%|███████▌  |                    \n",
-       "                             38/50 [09:00<03:00, 15.04s/it]#015                                                    \n",
-       "                             #015{'loss': '-0.0003246', 'grad_norm': '3.2', 'learning_rate':                       \n",
+       "
[06/26/26 11:11:39] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             74%|███████▍  | 37/50 [08:46<03:04, 14.16s/it]#015 76%|███████▌  |                    \n",
+       "                             38/50 [09:02<02:57, 14.76s/it]#015                                                    \n",
+       "                             #015{'loss': '-0.003074', 'grad_norm': '2.631', 'learning_rate':                      \n",
        "                             '2.6e-09', 'num_tokens': '2.457e+06', 'completions/mean_length':                      \n",
-       "                             '33.5', 'completions/min_length': '21', 'completions/max_length':                     \n",
+       "                             '33.31', 'completions/min_length': '21', 'completions/max_length':                    \n",
        "                             '62', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '33.5',                                         \n",
+       "                             'completions/mean_terminated_length': '33.31',                                        \n",
        "                             'completions/min_terminated_length': '21',                                            \n",
        "                             'completions/max_terminated_length': '62',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7188',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4514',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9151',                                               \n",
-       "                             'rewards/format_reward/std': '0.2784', 'reward': '1.176',                             \n",
-       "                             'reward_std': '0.5335', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '9.85e-05', 'entropy': '0.02256', 'clip_ratio/low_mean': '0',                         \n",
+       "                             'rewards/tool_call_reward/mean': '0.7109',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4551',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9135',                                               \n",
+       "                             'rewards/format_reward/std': '0.2831', 'reward': '1.168',                             \n",
+       "                             'reward_std': '0.5377', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '4.227e-05', 'entropy': '0.0207', 'clip_ratio/low_mean': '0',                         \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '16.36', 'epoch': '0.304'}                                               \n",
+       "                             'step_time': '16.14', 'epoch': '0.304'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:39:17]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027803;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027804;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m74\u001b[0m%|███████▍  | \u001b[1m37\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m08:44<03:08, \u001b[1m14.\u001b[0m47s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m76\u001b[0m%|███████▌  |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m38\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:00<03:00, \u001b[1m15.\u001b[0m04s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.0003246', 'grad_norm': '3.2', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:11:39]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029507;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029508;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m74\u001b[0m%|███████▍  | \u001b[1m37\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m08:46<03:04, \u001b[1m14.\u001b[0m16s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m76\u001b[0m%|███████▌  |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m38\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:02<02:57, \u001b[1m14.\u001b[0m76s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.003074', 'grad_norm': '2.631', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '2.6e-09', 'num_tokens': '2.457e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '33.5', 'completions/min_length': '21', 'completions/max_length':   \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '33.31', 'completions/min_length': '21', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '62', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '33.5',                       \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '33.31',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '21',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '62',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7188',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4514',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9151',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2784', 'reward': '1.176',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5335', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '9.85e-05', 'entropy': '0.02256', 'clip_ratio/low_mean': '0',       \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7109',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4551',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9135',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2831', 'reward': '1.168',           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.5377', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '4.227e-05', 'entropy': '0.0207', 'clip_ratio/low_mean': '0',       \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '16.36', 'epoch': '0.304'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '16.14', 'epoch': '0.304'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -5806,49 +5795,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:39:33] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             76%|███████▌  | 38/50 [09:00<03:00, 15.04s/it]#015 78%|███████▊  |                    \n",
-       "                             39/50 [09:16<02:47, 15.26s/it]#015                                                    \n",
-       "                             #015{'loss': '0.006804', 'grad_norm': '1.223', 'learning_rate':                       \n",
+       "
[06/26/26 11:11:54] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             76%|███████▌  | 38/50 [09:02<02:57, 14.76s/it]#015 78%|███████▊  |                    \n",
+       "                             39/50 [09:18<02:46, 15.15s/it]#015                                                    \n",
+       "                             #015{'loss': '0.002493', 'grad_norm': '2.959', 'learning_rate':                       \n",
        "                             '2.4e-09', 'num_tokens': '2.522e+06', 'completions/mean_length':                      \n",
-       "                             '42.23', 'completions/min_length': '22', 'completions/max_length':                    \n",
+       "                             '41.82', 'completions/min_length': '22', 'completions/max_length':                    \n",
        "                             '128', 'completions/clipped_ratio': '0.0625',                                         \n",
-       "                             'completions/mean_terminated_length': '36.52',                                        \n",
+       "                             'completions/mean_terminated_length': '36.08',                                        \n",
        "                             'completions/min_terminated_length': '22',                                            \n",
        "                             'completions/max_terminated_length': '60',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7734',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4203',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9528',                                               \n",
-       "                             'rewards/format_reward/std': '0.2136', 'reward': '1.25',                              \n",
-       "                             'reward_std': '0.4741', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '2.231e-05', 'entropy': '0.01186', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'rewards/tool_call_reward/mean': '0.7812',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.415',                                              \n",
+       "                             'rewards/format_reward/mean': '0.9371',                                               \n",
+       "                             'rewards/format_reward/std': '0.2445', 'reward': '1.25',                              \n",
+       "                             'reward_std': '0.4865', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '3.988e-05', 'entropy': '0.01252', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '15.77', 'epoch': '0.312'}                                               \n",
+       "                             'step_time': '16.04', 'epoch': '0.312'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:39:33]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027811;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027812;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m76\u001b[0m%|███████▌  | \u001b[1m38\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:00<03:00, \u001b[1m15.\u001b[0m04s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m78\u001b[0m%|███████▊  |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m39\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:16<02:47, \u001b[1m15.\u001b[0m26s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.006804', 'grad_norm': '1.223', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:11:54]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029515;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029516;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m76\u001b[0m%|███████▌  | \u001b[1m38\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:02<02:57, \u001b[1m14.\u001b[0m76s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m78\u001b[0m%|███████▊  |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m39\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:18<02:46, \u001b[1m15.\u001b[0m15s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.002493', 'grad_norm': '2.959', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '2.4e-09', 'num_tokens': '2.522e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '42.23', 'completions/min_length': '22', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '41.82', 'completions/min_length': '22', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '128', 'completions/clipped_ratio': '0.0625',                       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '36.52',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '36.08',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '22',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '60',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7734',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4203',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9528',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2136', 'reward': '1.25',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.4741', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '2.231e-05', 'entropy': '0.01186', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7812',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.415',                            \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9371',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2445', 'reward': '1.25',            \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.4865', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '3.988e-05', 'entropy': '0.01252', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '15.77', 'epoch': '0.312'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '16.04', 'epoch': '0.312'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -5857,49 +5846,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:39:43] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             78%|███████▊  | 39/50 [09:16<02:47, 15.26s/it]#015 80%|████████  |                    \n",
-       "                             40/50 [09:28<02:23, 14.38s/it]#015                                                    \n",
-       "                             #015{'loss': '0.002718', 'grad_norm': '2.099', 'learning_rate':                       \n",
+       "
[06/26/26 11:12:05] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             78%|███████▊  | 39/50 [09:18<02:46, 15.15s/it]#015 80%|████████  |                    \n",
+       "                             40/50 [09:30<02:23, 14.38s/it]#015                                                    \n",
+       "                             #015{'loss': '0.007042', 'grad_norm': '3.101', 'learning_rate':                       \n",
        "                             '2.2e-09', 'num_tokens': '2.585e+06', 'completions/mean_length':                      \n",
-       "                             '36.89', 'completions/min_length': '22', 'completions/max_length':                    \n",
+       "                             '36.74', 'completions/min_length': '22', 'completions/max_length':                    \n",
        "                             '59', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '36.89',                                        \n",
+       "                             'completions/mean_terminated_length': '36.74',                                        \n",
        "                             'completions/min_terminated_length': '22',                                            \n",
        "                             'completions/max_terminated_length': '59',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7812',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.415',                                              \n",
-       "                             'rewards/format_reward/mean': '0.945', 'rewards/format_reward/std':                   \n",
-       "                             '0.2296', 'reward': '1.254', 'reward_std': '0.4783',                                  \n",
-       "                             'frac_reward_zero_std': '0', 'kl': '0.0003581', 'entropy':                            \n",
-       "                             '0.1239', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',                      \n",
-       "                             'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',                              \n",
-       "                             'clip_ratio/region_mean': '0', 'step_time': '12.31', 'epoch':                         \n",
-       "                             '0.32'}                                                                               \n",
+       "                             'rewards/tool_call_reward/mean': '0.8281',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.3788',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9607',                                               \n",
+       "                             'rewards/format_reward/std': '0.1955', 'reward': '1.308',                             \n",
+       "                             'reward_std': '0.431', 'frac_reward_zero_std': '0', 'kl':                             \n",
+       "                             '0.0004843', 'entropy': '0.1136', 'clip_ratio/low_mean': '0',                         \n",
+       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
+       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
+       "                             'step_time': '12.57', 'epoch': '0.32'}                                                \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:39:43]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027819;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027820;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m78\u001b[0m%|███████▊  | \u001b[1m39\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:16<02:47, \u001b[1m15.\u001b[0m26s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m80\u001b[0m%|████████  |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m40\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:28<02:23, \u001b[1m14.\u001b[0m38s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.002718', 'grad_norm': '2.099', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:12:05]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029523;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029524;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m78\u001b[0m%|███████▊  | \u001b[1m39\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:18<02:46, \u001b[1m15.\u001b[0m15s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m80\u001b[0m%|████████  |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m40\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:30<02:23, \u001b[1m14.\u001b[0m38s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.007042', 'grad_norm': '3.101', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '2.2e-09', 'num_tokens': '2.585e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '36.89', 'completions/min_length': '22', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '36.74', 'completions/min_length': '22', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '59', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '36.89',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '36.74',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '22',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '59',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7812',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.415',                            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.945', 'rewards/format_reward/std': \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.2296', 'reward': '1.254', 'reward_std': '0.4783',                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'frac_reward_zero_std': '0', 'kl': '0.0003581', 'entropy':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.1239', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/region_mean': '0', 'step_time': '12.31', 'epoch':       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.32'\u001b[1m}\u001b[0m                                                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8281',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.3788',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9607',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.1955', 'reward': '1.308',           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.431', 'frac_reward_zero_std': '0', 'kl':           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.0004843', 'entropy': '0.1136', 'clip_ratio/low_mean': '0',       \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'step_time': '12.57', 'epoch': '0.32'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -5908,49 +5897,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:40:00] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             80%|████████  | 40/50 [09:28<02:23, 14.38s/it]#015 82%|████████▏ |                    \n",
-       "                             41/50 [09:45<02:15, 15.03s/it]#015                                                    \n",
-       "                             #015{'loss': '0.001328', 'grad_norm': '2.035', 'learning_rate':                       \n",
-       "                             '2e-09', 'num_tokens': '2.647e+06', 'completions/mean_length':                        \n",
-       "                             '31.13', 'completions/min_length': '19', 'completions/max_length':                    \n",
+       "
[06/26/26 11:12:21] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             80%|████████  | 40/50 [09:30<02:23, 14.38s/it]#015 82%|████████▏ |                    \n",
+       "                             41/50 [09:46<02:12, 14.71s/it]#015                                                    \n",
+       "                             #015{'loss': '0.001328', 'grad_norm': '1.698', 'learning_rate':                       \n",
+       "                             '2e-09', 'num_tokens': '2.646e+06', 'completions/mean_length':                        \n",
+       "                             '31.11', 'completions/min_length': '19', 'completions/max_length':                    \n",
        "                             '77', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '31.13',                                        \n",
+       "                             'completions/mean_terminated_length': '31.11',                                        \n",
        "                             'completions/min_terminated_length': '19',                                            \n",
        "                             'completions/max_terminated_length': '77',                                            \n",
        "                             'rewards/tool_call_reward/mean': '0.7734',                                            \n",
        "                             'rewards/tool_call_reward/std': '0.4203',                                             \n",
        "                             'rewards/format_reward/mean': '0.9058',                                               \n",
-       "                             'rewards/format_reward/std': '0.2939', 'reward': '1.226',                             \n",
+       "                             'rewards/format_reward/std': '0.2941', 'reward': '1.226',                             \n",
        "                             'reward_std': '0.5212', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '4.548e-05', 'entropy': '0.016', 'clip_ratio/low_mean': '0',                          \n",
+       "                             '2.695e-05', 'entropy': '0.01587', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '16.54', 'epoch': '0.328'}                                               \n",
+       "                             'step_time': '15.47', 'epoch': '0.328'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:40:00]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027827;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027828;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m80\u001b[0m%|████████  | \u001b[1m40\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:28<02:23, \u001b[1m14.\u001b[0m38s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m82\u001b[0m%|████████▏ |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m41\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:45<02:15, \u001b[1m15.\u001b[0m03s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.001328', 'grad_norm': '2.035', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '2e-09', 'num_tokens': '2.647e+06', 'completions/mean_length':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '31.13', 'completions/min_length': '19', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:12:21]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029531;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029532;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m80\u001b[0m%|████████  | \u001b[1m40\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:30<02:23, \u001b[1m14.\u001b[0m38s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m82\u001b[0m%|████████▏ |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m41\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:46<02:12, \u001b[1m14.\u001b[0m71s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.001328', 'grad_norm': '1.698', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '2e-09', 'num_tokens': '2.646e+06', 'completions/mean_length':      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '31.11', 'completions/min_length': '19', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '77', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '31.13',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '31.11',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '19',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '77',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7734',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4203',                           \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9058',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2939', 'reward': '1.226',           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2941', 'reward': '1.226',           \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'reward_std': '0.5212', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '4.548e-05', 'entropy': '0.016', 'clip_ratio/low_mean': '0',        \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '2.695e-05', 'entropy': '0.01587', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '16.54', 'epoch': '0.328'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '15.47', 'epoch': '0.328'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -5959,49 +5948,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:40:14] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             82%|████████▏ | 41/50 [09:45<02:15, 15.03s/it]#015 84%|████████▍ |                    \n",
-       "                             42/50 [09:57<01:54, 14.35s/it]#015                                                    \n",
-       "                             #015{'loss': '-0.003071', 'grad_norm': '3.307', 'learning_rate':                      \n",
+       "
[06/26/26 11:12:32] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             82%|████████▏ | 41/50 [09:46<02:12, 14.71s/it]#015 84%|████████▍ |                    \n",
+       "                             42/50 [09:58<01:50, 13.84s/it]#015                                                    \n",
+       "                             #015{'loss': '-0.007931', 'grad_norm': '2.842', 'learning_rate':                      \n",
        "                             '1.8e-09', 'num_tokens': '2.724e+06', 'completions/mean_length':                      \n",
-       "                             '32.88', 'completions/min_length': '22', 'completions/max_length':                    \n",
+       "                             '32.8', 'completions/min_length': '22', 'completions/max_length':                     \n",
        "                             '45', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '32.88',                                        \n",
+       "                             'completions/mean_terminated_length': '32.8',                                         \n",
        "                             'completions/min_terminated_length': '22',                                            \n",
        "                             'completions/max_terminated_length': '45',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.8516',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.3569',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9607',                                               \n",
-       "                             'rewards/format_reward/std': '0.1957', 'reward': '1.332',                             \n",
-       "                             'reward_std': '0.4132', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.000215', 'entropy': '0.02876', 'clip_ratio/low_mean': '0',                         \n",
+       "                             'rewards/tool_call_reward/mean': '0.8281',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.3788',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9371',                                               \n",
+       "                             'rewards/format_reward/std': '0.2445', 'reward': '1.297',                             \n",
+       "                             'reward_std': '0.4592', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '0.0001043', 'entropy': '0.02917', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '12.75', 'epoch': '0.336'}                                               \n",
+       "                             'step_time': '11.8', 'epoch': '0.336'}                                                \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:40:14]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027835;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027836;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m82\u001b[0m%|████████▏ | \u001b[1m41\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:45<02:15, \u001b[1m15.\u001b[0m03s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m84\u001b[0m%|████████▍ |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m42\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:57<01:54, \u001b[1m14.\u001b[0m35s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.003071', 'grad_norm': '3.307', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:12:32]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029539;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029540;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m82\u001b[0m%|████████▏ | \u001b[1m41\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:46<02:12, \u001b[1m14.\u001b[0m71s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m84\u001b[0m%|████████▍ |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m42\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:58<01:50, \u001b[1m13.\u001b[0m84s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.007931', 'grad_norm': '2.842', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '1.8e-09', 'num_tokens': '2.724e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '32.88', 'completions/min_length': '22', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '32.8', 'completions/min_length': '22', 'completions/max_length':   \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '45', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '32.88',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '32.8',                       \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '22',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '45',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8516',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.3569',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9607',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.1957', 'reward': '1.332',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.4132', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.000215', 'entropy': '0.02876', 'clip_ratio/low_mean': '0',       \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8281',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.3788',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9371',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2445', 'reward': '1.297',           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.4592', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.0001043', 'entropy': '0.02917', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '12.75', 'epoch': '0.336'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '11.8', 'epoch': '0.336'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -6010,49 +5999,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:40:25] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             84%|████████▍ | 42/50 [09:57<01:54, 14.35s/it]#015 86%|████████▌ |                    \n",
-       "                             43/50 [10:09<01:34, 13.51s/it]#015                                                    \n",
-       "                             #015{'loss': '0.003902', 'grad_norm': '3.486', 'learning_rate':                       \n",
-       "                             '1.6e-09', 'num_tokens': '2.784e+06', 'completions/mean_length':                      \n",
-       "                             '26.62', 'completions/min_length': '20', 'completions/max_length':                    \n",
-       "                             '42', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '26.62',                                        \n",
+       "
[06/26/26 11:12:48] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             84%|████████▍ | 42/50 [09:58<01:50, 13.84s/it]#015 86%|████████▌ |                    \n",
+       "                             43/50 [10:10<01:32, 13.24s/it]#015                                                    \n",
+       "                             #015{'loss': '-0.00114', 'grad_norm': '3.581', 'learning_rate':                       \n",
+       "                             '1.6e-09', 'num_tokens': '2.783e+06', 'completions/mean_length':                      \n",
+       "                             '26.7', 'completions/min_length': '20', 'completions/max_length':                     \n",
+       "                             '48', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '26.7',                                         \n",
        "                             'completions/min_terminated_length': '20',                                            \n",
-       "                             'completions/max_terminated_length': '42',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7734',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4203',                                             \n",
+       "                             'completions/max_terminated_length': '48',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.7656',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4253',                                             \n",
        "                             'rewards/format_reward/mean': '0.9293',                                               \n",
-       "                             'rewards/format_reward/std': '0.258', 'reward': '1.238',                              \n",
-       "                             'reward_std': '0.4983', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.0002812', 'entropy': '0.01698', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'rewards/format_reward/std': '0.258', 'reward': '1.23',                               \n",
+       "                             'reward_std': '0.502', 'frac_reward_zero_std': '0', 'kl':                             \n",
+       "                             '0.0001331', 'entropy': '0.01795', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '11.52', 'epoch': '0.344'}                                               \n",
+       "                             'step_time': '11.82', 'epoch': '0.344'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:40:25]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027843;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027844;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m84\u001b[0m%|████████▍ | \u001b[1m42\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:57<01:54, \u001b[1m14.\u001b[0m35s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m86\u001b[0m%|████████▌ |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m43\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:09<01:34, \u001b[1m13.\u001b[0m51s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.003902', 'grad_norm': '3.486', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '1.6e-09', 'num_tokens': '2.784e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '26.62', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '42', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '26.62',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:12:48]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029547;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029548;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m84\u001b[0m%|████████▍ | \u001b[1m42\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:58<01:50, \u001b[1m13.\u001b[0m84s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m86\u001b[0m%|████████▌ |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m43\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:10<01:32, \u001b[1m13.\u001b[0m24s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.00114', 'grad_norm': '3.581', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '1.6e-09', 'num_tokens': '2.783e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '26.7', 'completions/min_length': '20', 'completions/max_length':   \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '48', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '26.7',                       \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '42',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7734',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4203',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '48',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7656',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4253',                           \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9293',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.258', 'reward': '1.238',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.4983', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0002812', 'entropy': '0.01698', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.258', 'reward': '1.23',             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.502', 'frac_reward_zero_std': '0', 'kl':           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.0001331', 'entropy': '0.01795', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '11.52', 'epoch': '0.344'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '11.82', 'epoch': '0.344'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -6061,49 +6050,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:40:41] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             86%|████████▌ | 43/50 [10:09<01:34, 13.51s/it]#015 88%|████████▊ |                    \n",
-       "                             44/50 [10:24<01:23, 13.88s/it]#015                                                    \n",
-       "                             #015{'loss': '-0.001483', 'grad_norm': '1.746', 'learning_rate':                      \n",
+       "
[06/26/26 11:12:59] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             86%|████████▌ | 43/50 [10:10<01:32, 13.24s/it]#015 88%|████████▊ |                    \n",
+       "                             44/50 [10:24<01:21, 13.65s/it]#015                                                    \n",
+       "                             #015{'loss': '0.002971', 'grad_norm': '2.827', 'learning_rate':                       \n",
        "                             '1.4e-09', 'num_tokens': '2.851e+06', 'completions/mean_length':                      \n",
-       "                             '28.79', 'completions/min_length': '21', 'completions/max_length':                    \n",
+       "                             '28.79', 'completions/min_length': '19', 'completions/max_length':                    \n",
        "                             '40', 'completions/clipped_ratio': '0',                                               \n",
        "                             'completions/mean_terminated_length': '28.79',                                        \n",
-       "                             'completions/min_terminated_length': '21',                                            \n",
+       "                             'completions/min_terminated_length': '19',                                            \n",
        "                             'completions/max_terminated_length': '40',                                            \n",
        "                             'rewards/tool_call_reward/mean': '0.7266',                                            \n",
        "                             'rewards/tool_call_reward/std': '0.4475',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9451',                                               \n",
-       "                             'rewards/format_reward/std': '0.2294', 'reward': '1.199',                             \n",
-       "                             'reward_std': '0.5036', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.0003025', 'entropy': '0.03641', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'rewards/format_reward/mean': '0.9372',                                               \n",
+       "                             'rewards/format_reward/std': '0.2442', 'reward': '1.195',                             \n",
+       "                             'reward_std': '0.511', 'frac_reward_zero_std': '0', 'kl':                             \n",
+       "                             '0.0001698', 'entropy': '0.03675', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '14.75', 'epoch': '0.352'}                                               \n",
+       "                             'step_time': '14.59', 'epoch': '0.352'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:40:41]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027851;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027852;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m86\u001b[0m%|████████▌ | \u001b[1m43\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:09<01:34, \u001b[1m13.\u001b[0m51s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m88\u001b[0m%|████████▊ |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m44\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:24<01:23, \u001b[1m13.\u001b[0m88s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.001483', 'grad_norm': '1.746', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:12:59]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029555;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029556;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m86\u001b[0m%|████████▌ | \u001b[1m43\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:10<01:32, \u001b[1m13.\u001b[0m24s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m88\u001b[0m%|████████▊ |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m44\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:24<01:21, \u001b[1m13.\u001b[0m65s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.002971', 'grad_norm': '2.827', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '1.4e-09', 'num_tokens': '2.851e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '28.79', 'completions/min_length': '21', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '28.79', 'completions/min_length': '19', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '40', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '28.79',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '21',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '19',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '40',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7266',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4475',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9451',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2294', 'reward': '1.199',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5036', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0003025', 'entropy': '0.03641', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9372',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2442', 'reward': '1.195',           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.511', 'frac_reward_zero_std': '0', 'kl':           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.0001698', 'entropy': '0.03675', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '14.75', 'epoch': '0.352'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '14.59', 'epoch': '0.352'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -6112,49 +6101,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:40:51] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             88%|████████▊ | 44/50 [10:24<01:23, 13.88s/it]#015 90%|█████████ |                    \n",
-       "                             45/50 [10:35<01:06, 13.20s/it]#015                                                    \n",
-       "                             #015{'loss': '-0.005117', 'grad_norm': '3.015', 'learning_rate':                      \n",
+       "
[06/26/26 11:13:15] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             88%|████████▊ | 44/50 [10:24<01:21, 13.65s/it]#015 90%|█████████ |                    \n",
+       "                             45/50 [10:36<01:05, 13.02s/it]#015                                                    \n",
+       "                             #015{'loss': '0.001591', 'grad_norm': '3.238', 'learning_rate':                       \n",
        "                             '1.2e-09', 'num_tokens': '2.911e+06', 'completions/mean_length':                      \n",
-       "                             '32.46', 'completions/min_length': '21', 'completions/max_length':                    \n",
-       "                             '44', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '32.46',                                        \n",
+       "                             '32.55', 'completions/min_length': '21', 'completions/max_length':                    \n",
+       "                             '45', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '32.55',                                        \n",
        "                             'completions/min_terminated_length': '21',                                            \n",
-       "                             'completions/max_terminated_length': '44',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.8828',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.3229',                                             \n",
-       "                             'rewards/format_reward/mean': '0.8979',                                               \n",
-       "                             'rewards/format_reward/std': '0.3049', 'reward': '1.332',                             \n",
-       "                             'reward_std': '0.4673', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '4.154e-05', 'entropy': '0.009141', 'clip_ratio/low_mean': '0',                       \n",
+       "                             'completions/max_terminated_length': '45',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.8984',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.3033',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9136',                                               \n",
+       "                             'rewards/format_reward/std': '0.2829', 'reward': '1.355',                             \n",
+       "                             'reward_std': '0.4362', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '5.299e-06', 'entropy': '0.009558', 'clip_ratio/low_mean': '0',                       \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '11.61', 'epoch': '0.36'}                                                \n",
+       "                             'step_time': '11.53', 'epoch': '0.36'}                                                \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:40:51]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027859;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027860;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m88\u001b[0m%|████████▊ | \u001b[1m44\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:24<01:23, \u001b[1m13.\u001b[0m88s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m90\u001b[0m%|█████████ |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m45\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:35<01:06, \u001b[1m13.\u001b[0m20s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.005117', 'grad_norm': '3.015', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:13:15]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029563;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029564;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m88\u001b[0m%|████████▊ | \u001b[1m44\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:24<01:21, \u001b[1m13.\u001b[0m65s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m90\u001b[0m%|█████████ |  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m45\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:36<01:05, \u001b[1m13.\u001b[0m02s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.001591', 'grad_norm': '3.238', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '1.2e-09', 'num_tokens': '2.911e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '32.46', 'completions/min_length': '21', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '44', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '32.46',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '32.55', 'completions/min_length': '21', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '45', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '32.55',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '21',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '44',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8828',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.3229',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8979',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.3049', 'reward': '1.332',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.4673', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '4.154e-05', 'entropy': '0.009141', 'clip_ratio/low_mean': '0',     \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '45',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8984',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.3033',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9136',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2829', 'reward': '1.355',           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.4362', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '5.299e-06', 'entropy': '0.009558', 'clip_ratio/low_mean': '0',     \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '11.61', 'epoch': '0.36'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '11.53', 'epoch': '0.36'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -6163,49 +6152,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:41:07] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             90%|█████████ | 45/50 [10:35<01:06, 13.20s/it]#015 92%|█████████▏|                    \n",
-       "                             46/50 [10:48<00:52, 13.04s/it]#015                                                    \n",
-       "                             #015{'loss': '0.01022', 'grad_norm': '3.074', 'learning_rate':                        \n",
-       "                             '1e-09', 'num_tokens': '2.974e+06', 'completions/mean_length':                        \n",
-       "                             '34.24', 'completions/min_length': '21', 'completions/max_length':                    \n",
-       "                             '67', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '34.24',                                        \n",
+       "
[06/26/26 11:13:26] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             90%|█████████ | 45/50 [10:36<01:05, 13.02s/it]#015 92%|█████████▏|                    \n",
+       "                             46/50 [10:48<00:51, 12.95s/it]#015                                                    \n",
+       "                             #015{'loss': '0.009148', 'grad_norm': '2.514', 'learning_rate':                       \n",
+       "                             '1e-09', 'num_tokens': '2.973e+06', 'completions/mean_length':                        \n",
+       "                             '34.28', 'completions/min_length': '21', 'completions/max_length':                    \n",
+       "                             '69', 'completions/clipped_ratio': '0',                                               \n",
+       "                             'completions/mean_terminated_length': '34.28',                                        \n",
        "                             'completions/min_terminated_length': '21',                                            \n",
-       "                             'completions/max_terminated_length': '67',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.5703',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.497',                                              \n",
-       "                             'rewards/format_reward/mean': '0.9293',                                               \n",
-       "                             'rewards/format_reward/std': '0.258', 'reward': '1.035',                              \n",
-       "                             'reward_std': '0.5516', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.0002523', 'entropy': '0.0557', 'clip_ratio/low_mean': '0',                         \n",
+       "                             'completions/max_terminated_length': '69',                                            \n",
+       "                             'rewards/tool_call_reward/mean': '0.6328',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4839',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9529',                                               \n",
+       "                             'rewards/format_reward/std': '0.2133', 'reward': '1.109',                             \n",
+       "                             'reward_std': '0.525', 'frac_reward_zero_std': '0', 'kl':                             \n",
+       "                             '0.0002607', 'entropy': '0.0535', 'clip_ratio/low_mean': '0',                         \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '12.63', 'epoch': '0.368'}                                               \n",
+       "                             'step_time': '12.78', 'epoch': '0.368'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:41:07]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027867;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027868;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m90\u001b[0m%|█████████ | \u001b[1m45\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:35<01:06, \u001b[1m13.\u001b[0m20s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m92\u001b[0m%|█████████▏|  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m46\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:48<00:52, \u001b[1m13.\u001b[0m04s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.01022', 'grad_norm': '3.074', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '1e-09', 'num_tokens': '2.974e+06', 'completions/mean_length':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '34.24', 'completions/min_length': '21', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '67', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '34.24',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:13:26]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029571;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029572;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m90\u001b[0m%|█████████ | \u001b[1m45\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:36<01:05, \u001b[1m13.\u001b[0m02s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m92\u001b[0m%|█████████▏|  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m46\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:48<00:51, \u001b[1m12.\u001b[0m95s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.009148', 'grad_norm': '2.514', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '1e-09', 'num_tokens': '2.973e+06', 'completions/mean_length':      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '34.28', 'completions/min_length': '21', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '69', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '34.28',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '21',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '67',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.5703',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.497',                            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9293',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.258', 'reward': '1.035',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5516', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0002523', 'entropy': '0.0557', 'clip_ratio/low_mean': '0',       \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '69',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.6328',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4839',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9529',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2133', 'reward': '1.109',           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.525', 'frac_reward_zero_std': '0', 'kl':           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.0002607', 'entropy': '0.0535', 'clip_ratio/low_mean': '0',       \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '12.63', 'epoch': '0.368'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '12.78', 'epoch': '0.368'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -6214,49 +6203,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:41:18] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             92%|█████████▏| 46/50 [10:48<00:52, 13.04s/it]#015 94%|█████████▍|                    \n",
-       "                             47/50 [11:03<00:40, 13.49s/it]#015                                                    \n",
-       "                             #015{'loss': '-0.005655', 'grad_norm': '2.675', 'learning_rate':                      \n",
+       "
[06/26/26 11:13:42] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             92%|█████████▏| 46/50 [10:48<00:51, 12.95s/it]#015 94%|█████████▍|                    \n",
+       "                             47/50 [11:03<00:40, 13.34s/it]#015                                                    \n",
+       "                             #015{'loss': '0.003021', 'grad_norm': '2.457', 'learning_rate':                       \n",
        "                             '8e-10', 'num_tokens': '3.033e+06', 'completions/mean_length':                        \n",
-       "                             '33.34', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '33.32', 'completions/min_length': '20', 'completions/max_length':                    \n",
        "                             '55', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '33.34',                                        \n",
+       "                             'completions/mean_terminated_length': '33.32',                                        \n",
        "                             'completions/min_terminated_length': '20',                                            \n",
        "                             'completions/max_terminated_length': '55',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7578',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4301',                                             \n",
-       "                             'rewards/format_reward/mean': '0.8978',                                               \n",
-       "                             'rewards/format_reward/std': '0.3051', 'reward': '1.207',                             \n",
-       "                             'reward_std': '0.5351', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '5.591e-05', 'entropy': '0.01559', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'rewards/tool_call_reward/mean': '0.7266',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4475',                                             \n",
+       "                             'rewards/format_reward/mean': '0.8743',                                               \n",
+       "                             'rewards/format_reward/std': '0.334', 'reward': '1.164',                              \n",
+       "                             'reward_std': '0.5659', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '1.138e-05', 'entropy': '0.01608', 'clip_ratio/low_mean': '0',                        \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '14.54', 'epoch': '0.376'}                                               \n",
+       "                             'step_time': '14.23', 'epoch': '0.376'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:41:18]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027875;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027876;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m92\u001b[0m%|█████████▏| \u001b[1m46\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:48<00:52, \u001b[1m13.\u001b[0m04s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m94\u001b[0m%|█████████▍|  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m47\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:03<00:40, \u001b[1m13.\u001b[0m49s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.005655', 'grad_norm': '2.675', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:13:42]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029579;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029580;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m92\u001b[0m%|█████████▏| \u001b[1m46\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:48<00:51, \u001b[1m12.\u001b[0m95s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m94\u001b[0m%|█████████▍|  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m47\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:03<00:40, \u001b[1m13.\u001b[0m34s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.003021', 'grad_norm': '2.457', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '8e-10', 'num_tokens': '3.033e+06', 'completions/mean_length':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '33.34', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '33.32', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '55', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '33.34',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '33.32',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '55',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7578',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4301',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8978',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.3051', 'reward': '1.207',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5351', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '5.591e-05', 'entropy': '0.01559', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7266',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4475',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8743',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.334', 'reward': '1.164',            \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.5659', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '1.138e-05', 'entropy': '0.01608', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '14.54', 'epoch': '0.376'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '14.23', 'epoch': '0.376'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -6265,49 +6254,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:41:34] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             94%|█████████▍| 47/50 [11:03<00:40, 13.49s/it]#015 96%|█████████▌|                    \n",
-       "                             48/50 [11:19<00:28, 14.30s/it]#015                                                    \n",
-       "                             #015{'loss': '0.004514', 'grad_norm': '1.322', 'learning_rate':                       \n",
+       "
[06/26/26 11:13:58] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             94%|█████████▍| 47/50 [11:03<00:40, 13.34s/it]#015 96%|█████████▌|                    \n",
+       "                             48/50 [11:19<00:28, 14.08s/it]#015                                                    \n",
+       "                             #015{'loss': '0.007249', 'grad_norm': '1.409', 'learning_rate':                       \n",
        "                             '6e-10', 'num_tokens': '3.098e+06', 'completions/mean_length':                        \n",
-       "                             '37.12', 'completions/min_length': '20', 'completions/max_length':                    \n",
-       "                             '128', 'completions/clipped_ratio': '0.05469',                                        \n",
-       "                             'completions/mean_terminated_length': '31.86',                                        \n",
+       "                             '37.45', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '128', 'completions/clipped_ratio': '0.0625',                                         \n",
+       "                             'completions/mean_terminated_length': '31.41',                                        \n",
        "                             'completions/min_terminated_length': '20',                                            \n",
-       "                             'completions/max_terminated_length': '102',                                           \n",
-       "                             'rewards/tool_call_reward/mean': '0.7656',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4253',                                             \n",
-       "                             'rewards/format_reward/mean': '0.8895',                                               \n",
-       "                             'rewards/format_reward/std': '0.3165', 'reward': '1.21',                              \n",
-       "                             'reward_std': '0.5396', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.0001637', 'entropy': '0.01912', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'completions/max_terminated_length': '100',                                           \n",
+       "                             'rewards/tool_call_reward/mean': '0.7422',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4391',                                             \n",
+       "                             'rewards/format_reward/mean': '0.8739',                                               \n",
+       "                             'rewards/format_reward/std': '0.3349', 'reward': '1.179',                             \n",
+       "                             'reward_std': '0.5614', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '4.697e-05', 'entropy': '0.0159', 'clip_ratio/low_mean': '0',                         \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '16.17', 'epoch': '0.384'}                                               \n",
+       "                             'step_time': '15.81', 'epoch': '0.384'}                                               \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:41:34]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027883;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027884;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m94\u001b[0m%|█████████▍| \u001b[1m47\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:03<00:40, \u001b[1m13.\u001b[0m49s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m96\u001b[0m%|█████████▌|  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m48\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:19<00:28, \u001b[1m14.\u001b[0m30s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.004514', 'grad_norm': '1.322', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:13:58]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029587;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029588;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m94\u001b[0m%|█████████▍| \u001b[1m47\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:03<00:40, \u001b[1m13.\u001b[0m34s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m96\u001b[0m%|█████████▌|  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m48\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:19<00:28, \u001b[1m14.\u001b[0m08s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.007249', 'grad_norm': '1.409', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '6e-10', 'num_tokens': '3.098e+06', 'completions/mean_length':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '37.12', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '128', 'completions/clipped_ratio': '0.05469',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '31.86',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '37.45', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '128', 'completions/clipped_ratio': '0.0625',                       \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '31.41',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '102',                         \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7656',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4253',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8895',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.3165', 'reward': '1.21',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5396', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0001637', 'entropy': '0.01912', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '100',                         \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7422',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4391',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8739',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.3349', 'reward': '1.179',           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.5614', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '4.697e-05', 'entropy': '0.0159', 'clip_ratio/low_mean': '0',       \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '16.17', 'epoch': '0.384'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '15.81', 'epoch': '0.384'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -6316,49 +6305,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:41:50] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             96%|█████████▌| 48/50 [11:19<00:28, 14.30s/it]#015 98%|█████████▊|                    \n",
-       "                             49/50 [11:34<00:14, 14.48s/it]#015                                                    \n",
-       "                             #015{'loss': '-0.009328', 'grad_norm': '1.974', 'learning_rate':                      \n",
+       "
[06/26/26 11:14:09] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             96%|█████████▌| 48/50 [11:19<00:28, 14.08s/it]#015 98%|█████████▊|                    \n",
+       "                             49/50 [11:35<00:14, 14.71s/it]#015                                                    \n",
+       "                             #015{'loss': '-0.02007', 'grad_norm': '2.001', 'learning_rate':                       \n",
        "                             '4e-10', 'num_tokens': '3.167e+06', 'completions/mean_length':                        \n",
-       "                             '36.41', 'completions/min_length': '21', 'completions/max_length':                    \n",
+       "                             '37.2', 'completions/min_length': '21', 'completions/max_length':                     \n",
        "                             '109', 'completions/clipped_ratio': '0',                                              \n",
-       "                             'completions/mean_terminated_length': '36.41',                                        \n",
+       "                             'completions/mean_terminated_length': '37.2',                                         \n",
        "                             'completions/min_terminated_length': '21',                                            \n",
        "                             'completions/max_terminated_length': '109',                                           \n",
-       "                             'rewards/tool_call_reward/mean': '0.6953',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4621',                                             \n",
-       "                             'rewards/format_reward/mean': '0.8583',                                               \n",
-       "                             'rewards/format_reward/std': '0.3518', 'reward': '1.124',                             \n",
-       "                             'reward_std': '0.5863', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '3.141e-05', 'entropy': '0.01278', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '14.88', 'epoch': '0.392'}                                               \n",
+       "                             'rewards/tool_call_reward/mean': '0.7344',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.4434',                                             \n",
+       "                             'rewards/format_reward/mean': '0.882', 'rewards/format_reward/std':                   \n",
+       "                             '0.3253', 'reward': '1.175', 'reward_std': '0.5572',                                  \n",
+       "                             'frac_reward_zero_std': '0', 'kl': '8.196e-05', 'entropy':                            \n",
+       "                             '0.0125', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',                      \n",
+       "                             'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',                              \n",
+       "                             'clip_ratio/region_mean': '0', 'step_time': '16.16', 'epoch':                         \n",
+       "                             '0.392'}                                                                              \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:41:50]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027891;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027892;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m96\u001b[0m%|█████████▌| \u001b[1m48\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:19<00:28, \u001b[1m14.\u001b[0m30s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m98\u001b[0m%|█████████▊|  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:34<00:14, \u001b[1m14.\u001b[0m48s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.009328', 'grad_norm': '1.974', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:14:09]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029595;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029596;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m96\u001b[0m%|█████████▌| \u001b[1m48\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:19<00:28, \u001b[1m14.\u001b[0m08s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m98\u001b[0m%|█████████▊|  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:35<00:14, \u001b[1m14.\u001b[0m71s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.02007', 'grad_norm': '2.001', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '4e-10', 'num_tokens': '3.167e+06', 'completions/mean_length':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '36.41', 'completions/min_length': '21', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '37.2', 'completions/min_length': '21', 'completions/max_length':   \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '109', 'completions/clipped_ratio': '0',                            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '36.41',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '37.2',                       \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '21',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '109',                         \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.6953',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4621',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8583',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.3518', 'reward': '1.124',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5863', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '3.141e-05', 'entropy': '0.01278', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '14.88', 'epoch': '0.392'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7344',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4434',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.882', 'rewards/format_reward/std': \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.3253', 'reward': '1.175', 'reward_std': '0.5572',                \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'frac_reward_zero_std': '0', 'kl': '8.196e-05', 'entropy':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.0125', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',    \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',            \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'clip_ratio/region_mean': '0', 'step_time': '16.16', 'epoch':       \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '0.392'\u001b[1m}\u001b[0m                                                            \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -6367,49 +6356,49 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:42:07] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             98%|█████████▊| 49/50 [11:34<00:14, 14.48s/it]#015100%|██████████|                    \n",
-       "                             50/50 [11:48<00:00, 14.52s/it]#015                                                    \n",
-       "                             #015{'loss': '-0.00214', 'grad_norm': '2.388', 'learning_rate':                       \n",
+       "
[06/26/26 11:14:25] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             98%|█████████▊| 49/50 [11:35<00:14, 14.71s/it]#015100%|██████████|                    \n",
+       "                             50/50 [11:50<00:00, 14.97s/it]#015                                                    \n",
+       "                             #015{'loss': '0.004886', 'grad_norm': '3.012', 'learning_rate':                       \n",
        "                             '2e-10', 'num_tokens': '3.236e+06', 'completions/mean_length':                        \n",
-       "                             '29.54', 'completions/min_length': '20', 'completions/max_length':                    \n",
+       "                             '29.63', 'completions/min_length': '20', 'completions/max_length':                    \n",
        "                             '46', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '29.54',                                        \n",
+       "                             'completions/mean_terminated_length': '29.63',                                        \n",
        "                             'completions/min_terminated_length': '20',                                            \n",
        "                             'completions/max_terminated_length': '46',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.8984',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.3033',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9293',                                               \n",
-       "                             'rewards/format_reward/std': '0.258', 'reward': '1.363',                              \n",
-       "                             'reward_std': '0.4155', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '8.118e-05', 'entropy': '0.01057', 'clip_ratio/low_mean': '0',                        \n",
+       "                             'rewards/tool_call_reward/mean': '0.9219',                                            \n",
+       "                             'rewards/tool_call_reward/std': '0.2694',                                             \n",
+       "                             'rewards/format_reward/mean': '0.9686',                                               \n",
+       "                             'rewards/format_reward/std': '0.1756', 'reward': '1.406',                             \n",
+       "                             'reward_std': '0.3309', 'frac_reward_zero_std': '0', 'kl':                            \n",
+       "                             '6.488e-05', 'entropy': '0.009783', 'clip_ratio/low_mean': '0',                       \n",
        "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
        "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '14.61', 'epoch': '0.4'}                                                 \n",
+       "                             'step_time': '15.57', 'epoch': '0.4'}                                                 \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:42:07]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027899;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027900;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m98\u001b[0m%|█████████▊| \u001b[1m49\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:34<00:14, \u001b[1m14.\u001b[0m48s/it\u001b[1m]\u001b[0m#\u001b[1m015100\u001b[0m%|██████████|  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m50\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:48<00:00, \u001b[1m14.\u001b[0m52s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.00214', 'grad_norm': '2.388', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:14:25]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029603;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029604;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m98\u001b[0m%|█████████▊| \u001b[1m49\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:35<00:14, \u001b[1m14.\u001b[0m71s/it\u001b[1m]\u001b[0m#\u001b[1m015100\u001b[0m%|██████████|  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m50\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:50<00:00, \u001b[1m14.\u001b[0m97s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.004886', 'grad_norm': '3.012', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '2e-10', 'num_tokens': '3.236e+06', 'completions/mean_length':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '29.54', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '29.63', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         '46', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '29.54',                      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '29.63',                      \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '46',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8984',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.3033',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9293',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.258', 'reward': '1.363',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.4155', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '8.118e-05', 'entropy': '0.01057', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.9219',                          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.2694',                           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9686',                             \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.1756', 'reward': '1.406',           \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'reward_std': '0.3309', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         '6.488e-05', 'entropy': '0.009783', 'clip_ratio/low_mean': '0',     \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '14.61', 'epoch': '0.4'\u001b[1m}\u001b[0m                               \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                    \u001b[0m         'step_time': '15.57', 'epoch': '0.4'\u001b[1m}\u001b[0m                               \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -6418,20 +6407,20 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             100%|██████████| 50/50 [11:48<00:00, 14.52s/it]#015                                   \n",
-       "                             #015{'train_runtime': '708.8', 'train_samples_per_second': '9.03',                    \n",
-       "                             'train_steps_per_second': '0.071', 'train_loss': '0.003869',                          \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             100%|██████████| 50/50 [11:50<00:00, 14.97s/it]#015                                   \n",
+       "                             #015{'train_runtime': '710.8', 'train_samples_per_second': '9.004',                   \n",
+       "                             'train_steps_per_second': '0.07', 'train_loss': '0.003163',                           \n",
        "                             'epoch': '0.4'}                                                                       \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027907;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027908;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m100\u001b[0m%|██████████| \u001b[1m50\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:48<00:00, \u001b[1m14.\u001b[0m52s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'train_runtime': '708.8', 'train_samples_per_second': '9.03',  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'train_steps_per_second': '0.071', 'train_loss': '0.003869',        \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029611;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029612;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m100\u001b[0m%|██████████| \u001b[1m50\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:50<00:00, \u001b[1m14.\u001b[0m97s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'train_runtime': '710.8', 'train_samples_per_second': '9.004', \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         'train_steps_per_second': '0.07', 'train_loss': '0.003163',         \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         'epoch': '0.4'\u001b[1m}\u001b[0m                                                     \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -6441,17 +6430,17 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
-       "                             100%|██████████| 50/50 [11:48<00:00, 14.52s/it]#015100%|██████████|                   \n",
-       "                             50/50 [11:48<00:00, 14.18s/it]                                                        \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
+       "                             100%|██████████| 50/50 [11:50<00:00, 14.97s/it]#015100%|██████████|                   \n",
+       "                             50/50 [11:50<00:00, 14.22s/it]                                                        \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027915;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027916;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m100\u001b[0m%|██████████| \u001b[1m50\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:48<00:00, \u001b[1m14.\u001b[0m52s/it\u001b[1m]\u001b[0m#\u001b[1m015100\u001b[0m%|██████████| \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m50\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:48<00:00, \u001b[1m14.\u001b[0m18s/it\u001b[1m]\u001b[0m                                      \u001b[2m                  \u001b[0m\n"
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029619;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029620;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m100\u001b[0m%|██████████| \u001b[1m50\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:50<00:00, \u001b[1m14.\u001b[0m97s/it\u001b[1m]\u001b[0m#\u001b[1m015100\u001b[0m%|██████████| \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m50\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:50<00:00, \u001b[1m14.\u001b[0m22s/it\u001b[1m]\u001b[0m                                      \u001b[2m                  \u001b[0m\n"
       ]
      },
      "metadata": {},
@@ -6460,21 +6449,21 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:42:12] INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
[06/26/26 11:14:36] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             Writing model shards:   0%|          | 0/1 [00:00<?,                                  \n",
        "                             ?it/s]#015Writing model shards: 100%|██████████| 1/1 [00:02<00:00,                    \n",
-       "                             2.70s/it]#015Writing model shards: 100%|██████████| 1/1                               \n",
-       "                             [00:02<00:00,  2.70s/it]                                                              \n",
+       "                             2.73s/it]#015Writing model shards: 100%|██████████| 1/1                               \n",
+       "                             [00:02<00:00,  2.73s/it]                                                              \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:42:12]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027923;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027924;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m[06/26/26 11:14:36]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029627;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029628;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         Writing model shards:   \u001b[1m0\u001b[0m%|          | \u001b[1m0\u001b[0m/\u001b[1m1\u001b[0m \u001b[1m[\u001b[0m00:00                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             ++ echo 'Training Container Execution Completed'                                      \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027931;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027932;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029635;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029636;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         ++ echo 'Training Container Execution Completed'                    \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -6500,14 +6489,14 @@
     {
      "data": {
       "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-081812-20260626101812/algo-1-17824623 resources.py:31439\n",
-       "                             09:                                                                                   \n",
+       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
+       "                             49:                                                                                   \n",
        "                             Training Container Execution Completed                                                \n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m081812\u001b[0m-\u001b[1m20260626101812\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824623\u001b[0m \u001b]8;id=17027939;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027940;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m09\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029643;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029644;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
        "\u001b[2m                    \u001b[0m         Training Container Execution Completed                              \u001b[2m                  \u001b[0m\n"
       ]
      },
@@ -6517,11 +6506,11 @@
     {
      "data": {
       "text/html": [
-       "
[06/26/26 10:42:44] INFO     Final Resource Status: Completed                                    resources.py:31442\n",
+       "
[06/26/26 11:14:57] INFO     Final Resource Status: Completed                                    resources.py:31442\n",
        "
\n"
       ],
       "text/plain": [
-       "\u001b[2m[06/26/26 10:42:44]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m Final Resource Status: \u001b[1mCompleted\u001b[0m                                    \u001b]8;id=17027947;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027948;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31442\u001b\\\u001b[2m31442\u001b[0m\u001b]8;;\u001b\\\n"
+       "\u001b[2m[06/26/26 11:14:57]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m Final Resource Status: \u001b[1mCompleted\u001b[0m                                    \u001b]8;id=17029651;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029652;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31442\u001b\\\u001b[2m31442\u001b[0m\u001b]8;;\u001b\\\n"
       ]
      },
      "metadata": {},
@@ -6542,6 +6531,7 @@
     "import time\n",
     "from sagemaker.train.model_trainer import ModelTrainer\n",
     "from sagemaker.train.configs import SourceCode, Compute, InputData, OutputDataConfig, StoppingCondition\n",
+    "from sagemaker.train.distributed import Torchrun\n",
     "\n",
     "DEBUG_NO_UPDATE = False\n",
     "\n",
@@ -6552,7 +6542,8 @@
     "    role=role,\n",
     "    training_image=TRAINING_IMAGE,\n",
     "    base_job_name=base_job_name,\n",
-    "    source_code=SourceCode(source_dir=\"src\", entry_script=\"launch.sh\"),\n",
+    "    source_code=SourceCode(source_dir=\"src\", entry_script=\"train.py\"),\n",
+    "    distributed=Torchrun(process_count_per_node=8),\n",
     "    compute=Compute(instance_type=INSTANCE_TYPE, instance_count=1, volume_size_in_gb=200),\n",
     "    stopping_condition=StoppingCondition(max_runtime_in_seconds=6 * 60 * 60),\n",
     "    output_data_config=OutputDataConfig(s3_output_path=f\"s3://{bucket}/xlam-grpo/output/\"),\n",

From a358e09bf17376260c802ef255be927664affa81 Mon Sep 17 00:00:00 2001
From: DWarez 
Date: Sat, 27 Jun 2026 12:54:57 +0200
Subject: [PATCH 4/5] change: notebook login as auth, better markdown cells

add: references

Signed-off-by: DWarez 
---
 .../grpo-llm-trl/sagemaker-notebook.ipynb     | 6275 +----------------
 1 file changed, 388 insertions(+), 5887 deletions(-)

diff --git a/docs/sagemaker/notebooks/sagemaker-sdk/grpo-llm-trl/sagemaker-notebook.ipynb b/docs/sagemaker/notebooks/sagemaker-sdk/grpo-llm-trl/sagemaker-notebook.ipynb
index 51353e2e8..d0b73408a 100644
--- a/docs/sagemaker/notebooks/sagemaker-sdk/grpo-llm-trl/sagemaker-notebook.ipynb
+++ b/docs/sagemaker/notebooks/sagemaker-sdk/grpo-llm-trl/sagemaker-notebook.ipynb
@@ -23,6 +23,30 @@
     "You need AWS credentials, a SageMaker execution role with S3 access, quota for `ml.p4d.24xlarge` training in `us-east-1`, and a Hugging Face token that can download the base model.\n"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "id": "c6f9a26c",
+   "metadata": {},
+   "source": [
+    "## What is GRPO?\n",
+    "\n",
+    "Group Relative Policy Optimization (GRPO) is an RL fine-tuning method for language models. DeepSeek-R1 made it widely known: DeepSeek-R1-Zero used GRPO with rule-based rewards, while DeepSeek-R1 added cold-start data and a broader multi-stage training recipe around the RL stage.\n",
+    "\n",
+    "The key idea is simple: for each prompt, sample several completions, score each one, and update the model based on how each completion did relative to the other completions from the same prompt. Unlike PPO-style RLHF setups, GRPO does not need a separate value model. The group itself provides the baseline.\n",
+    "\n",
+    "```text\n",
+    "prompt\n",
+    "  -> sample N completions from the current model\n",
+    "  -> score each completion with a reward function\n",
+    "  -> compare each reward to the group mean/std\n",
+    "  -> increase probability of better-than-group completions\n",
+    "  -> decrease probability of worse-than-group completions\n",
+    "  -> apply a KL penalty so the model does not drift too far from the reference model\n",
+    "```\n",
+    "\n",
+    "For reasoning models, rewards often check things like answer correctness and output format. In this notebook, the same pattern is used for tool calling: the model gets reward when it emits the correct tool name and arguments, plus a smaller formatting reward for valid `` JSON.\n"
+   ]
+  },
   {
    "cell_type": "markdown",
    "id": "5b77b8dd",
@@ -60,33 +84,78 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 1,
    "id": "e1a0499a",
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "HF token loaded\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
+      "  from .autonotebook import tqdm as notebook_tqdm\n"
+     ]
+    }
+   ],
    "source": [
-    "import os\n",
-    "from huggingface_hub import login, get_token\n",
+    "from huggingface_hub import get_token, notebook_login\n",
     "\n",
-    "# Uses the HF_TOKEN env var if set; otherwise opens an interactive login prompt.\n",
-    "if not os.environ.get(\"HF_TOKEN\"):\n",
-    "    login()\n",
-    "HF_TOKEN = get_token()\n",
-    "assert HF_TOKEN, \"No HF token found — set HF_TOKEN or run login()\"\n",
+    "if not (HF_TOKEN := get_token()):\n",
+    "    notebook_login()\n",
+    "    HF_TOKEN = get_token()\n",
+    "\n",
+    "assert HF_TOKEN, \"No HF token found - set HF_TOKEN or run notebook_login()\"\n",
     "print(\"HF token loaded\")"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 2,
    "id": "954730c2",
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "sagemaker.config INFO - Not applying SDK defaults from location: /Library/Application Support/sagemaker/config.yaml\n",
+      "sagemaker.config INFO - Not applying SDK defaults from location: /Users/dwarez/Library/Application Support/sagemaker/config.yaml\n"
+     ]
+    },
+    {
+     "data": {
+      "text/html": [
+       "
[06/27/26 12:50:38] INFO     Loading cached SSO token for hf                                          tokens.py:312\n",
+       "
\n"
+      ],
+      "text/plain": [
+       "\u001b[2m[06/27/26 12:50:38]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m Loading cached SSO token for hf                                          \u001b]8;id=8266086;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/botocore/tokens.py\u001b\\\u001b[2mtokens.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=8266087;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/botocore/tokens.py#312\u001b\\\u001b[2m312\u001b[0m\u001b]8;;\u001b\\\n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "region: us-east-1\n",
+      "bucket: sagemaker-us-east-1-754289655784\n",
+      "role:   arn:aws:iam::754289655784:role/aws-reserved/sso.amazonaws.com/AWSReservedSSO_HF-Sandbox-access_a9a3037b77bf6782\n"
+     ]
+    }
+   ],
    "source": [
     "import boto3\n",
     "from sagemaker.core.helper.session_helper import Session, get_execution_role\n",
     "\n",
-    "# The training image lives in us-east-1, so keep the job, the S3 bucket and the image in one region.\n",
     "REGION = \"us-east-1\"\n",
     "sess = Session(boto_session=boto3.Session(region_name=REGION))\n",
     "bucket = sess.default_bucket()\n",
@@ -109,20 +178,30 @@
     "## Configuration\n",
     "\n",
     "`MODEL_ID` is the base model. `INSTANCE_TYPE` controls the training hardware. `TRAINING_IMAGE` is the ECR image SageMaker runs.\n",
-    "\n",
-    "Keep `TRAINING_IMAGE`, the SageMaker job, and the S3 bucket in the same AWS region.\n"
+    "\n"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 3,
+   "id": "f54550f3-220b-438b-aeba-d593bdb44538",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "role = \"arn:aws:iam::754289655784:role/sagemaker_execution_role\""
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
    "id": "7280ac2e",
    "metadata": {},
    "outputs": [],
    "source": [
     "MODEL_ID = \"HuggingFaceTB/SmolLM3-3B\"      # SmolLM3, HF's 3B instruct model\n",
-    "INSTANCE_TYPE = \"ml.p4d.24xlarge\"        # 8xA100 40GB\n",
+    "INSTANCE_TYPE = \"ml.p4d.24xlarge\"          # 8 x A100 40GB\n",
     "\n",
+    "# TODO: Temporary review image; replace with the public tutorial image before publishing.\n",
     "TRAINING_IMAGE = \"754289655784.dkr.ecr.us-east-1.amazonaws.com/hf-trl-grpo:sagemaker-trl-dev-e63f67e\""
    ]
   },
@@ -142,17 +221,189 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 5,
    "id": "2690d04e",
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "data": {
+      "text/html": [
+       "
[06/27/26 12:51:04] INFO     HTTP Request: GET https://huggingface.co/api/agent-harnesses \"HTTP/1.1 _client.py:1025\n",
+       "                             200 OK\"                                                                               \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "\u001b[2m[06/27/26 12:51:04]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m HTTP Request: \u001b[1mGET\u001b[0m \u001b[4mhttps://huggingface.co/api/agent-harnesses\u001b[0m \"HTTP/1.1 \u001b]8;id=8266096;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py\u001b\\\u001b[2m_client.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=8266097;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py#1025\u001b\\\u001b[2m1025\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         200 OK\"                                                                \u001b[2m               \u001b[0m\n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
                    INFO     HTTP Request: HEAD                                                     _client.py:1025\n",
+       "                             https://huggingface.co/datasets/Salesforce/xlam-function-calling-60k/r                \n",
+       "                             esolve/main/README.md \"HTTP/1.1 200 OK\"                                               \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m HTTP Request: \u001b[1mHEAD\u001b[0m                                                     \u001b]8;id=8266104;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py\u001b\\\u001b[2m_client.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=8266105;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py#1025\u001b\\\u001b[2m1025\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[4mhttps://huggingface.co/datasets/Salesforce/xlam-function-calling-60k/r\u001b[0m \u001b[2m               \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[4mesolve/main/README.md\u001b[0m \"HTTP/1.1 200 OK\"                                \u001b[2m               \u001b[0m\n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
                    INFO     HTTP Request: HEAD                                                     _client.py:1025\n",
+       "                             https://huggingface.co/datasets/Salesforce/xlam-function-calling-60k/r                \n",
+       "                             esolve/26d14ebfe18b1f7b524bd39b404b50af5dc97866/xlam-function-calling-                \n",
+       "                             60k.py \"HTTP/1.1 404 Not Found\"                                                       \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m HTTP Request: \u001b[1mHEAD\u001b[0m                                                     \u001b]8;id=8266112;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py\u001b\\\u001b[2m_client.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=8266113;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py#1025\u001b\\\u001b[2m1025\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[4mhttps://huggingface.co/datasets/Salesforce/xlam-function-calling-60k/r\u001b[0m \u001b[2m               \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[4mesolve/26d14ebfe18b1f7b524bd39b404b50af5dc97866/xlam-function-calling-\u001b[0m \u001b[2m               \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[4m60k.py\u001b[0m \"HTTP/1.1 404 Not Found\"                                        \u001b[2m               \u001b[0m\n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
                    INFO     HTTP Request: HEAD                                                     _client.py:1025\n",
+       "                             https://s3.amazonaws.com/datasets.huggingface.co/datasets/datasets/Sal                \n",
+       "                             esforce/xlam-function-calling-60k/Salesforce/xlam-function-calling-60k                \n",
+       "                             .py \"HTTP/1.1 404 Not Found\"                                                          \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m HTTP Request: \u001b[1mHEAD\u001b[0m                                                     \u001b]8;id=8266120;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py\u001b\\\u001b[2m_client.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=8266121;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py#1025\u001b\\\u001b[2m1025\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[4mhttps://s3.amazonaws.com/datasets.huggingface.co/datasets/datasets/Sal\u001b[0m \u001b[2m               \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[4mesforce/xlam-function-calling-60k/Salesforce/xlam-function-calling-60k\u001b[0m \u001b[2m               \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[4m.py\u001b[0m \"HTTP/1.1 404 Not Found\"                                           \u001b[2m               \u001b[0m\n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
                    INFO     HTTP Request: GET                                                      _client.py:1025\n",
+       "                             https://huggingface.co/api/datasets/Salesforce/xlam-function-calling-6                \n",
+       "                             0k/revision/26d14ebfe18b1f7b524bd39b404b50af5dc97866 \"HTTP/1.1 200 OK\"                \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m HTTP Request: \u001b[1mGET\u001b[0m                                                      \u001b]8;id=8266128;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py\u001b\\\u001b[2m_client.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=8266129;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py#1025\u001b\\\u001b[2m1025\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[4mhttps://huggingface.co/api/datasets/Salesforce/xlam-function-calling-6\u001b[0m \u001b[2m               \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[4m0k/revision/26d14ebfe18b1f7b524bd39b404b50af5dc97866\u001b[0m \"HTTP/1.1 200 OK\" \u001b[2m               \u001b[0m\n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
[06/27/26 12:51:05] INFO     HTTP Request: HEAD                                                     _client.py:1025\n",
+       "                             https://huggingface.co/datasets/Salesforce/xlam-function-calling-60k/r                \n",
+       "                             esolve/26d14ebfe18b1f7b524bd39b404b50af5dc97866/.huggingface.yaml                     \n",
+       "                             \"HTTP/1.1 404 Not Found\"                                                              \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "\u001b[2m[06/27/26 12:51:05]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m HTTP Request: \u001b[1mHEAD\u001b[0m                                                     \u001b]8;id=8266136;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py\u001b\\\u001b[2m_client.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=8266137;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py#1025\u001b\\\u001b[2m1025\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[4mhttps://huggingface.co/datasets/Salesforce/xlam-function-calling-60k/r\u001b[0m \u001b[2m               \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[4mesolve/26d14ebfe18b1f7b524bd39b404b50af5dc97866/.huggingface.yaml\u001b[0m      \u001b[2m               \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \"HTTP/1.1 404 Not Found\"                                               \u001b[2m               \u001b[0m\n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
                    INFO     HTTP Request: GET                                                      _client.py:1025\n",
+       "                             https://datasets-server.huggingface.co/info?dataset=Salesforce/xlam-fu                \n",
+       "                             nction-calling-60k \"HTTP/1.1 200 OK\"                                                  \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m HTTP Request: \u001b[1mGET\u001b[0m                                                      \u001b]8;id=8266144;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py\u001b\\\u001b[2m_client.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=8266145;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py#1025\u001b\\\u001b[2m1025\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[4mhttps://datasets-server.huggingface.co/info?\u001b[0m\u001b[4mdataset\u001b[0m\u001b[4m=\u001b[0m\u001b[4mSalesforce\u001b[0m\u001b[4m/xlam-fu\u001b[0m \u001b[2m               \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[4mnction-calling-60k\u001b[0m \"HTTP/1.1 200 OK\"                                   \u001b[2m               \u001b[0m\n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
                    INFO     HTTP Request: GET                                                      _client.py:1025\n",
+       "                             https://huggingface.co/api/datasets/Salesforce/xlam-function-calling-6                \n",
+       "                             0k/tree/26d14ebfe18b1f7b524bd39b404b50af5dc97866?recursive=false&expan                \n",
+       "                             d=false \"HTTP/1.1 200 OK\"                                                             \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m HTTP Request: \u001b[1mGET\u001b[0m                                                      \u001b]8;id=8266152;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py\u001b\\\u001b[2m_client.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=8266153;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py#1025\u001b\\\u001b[2m1025\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[4mhttps://huggingface.co/api/datasets/Salesforce/xlam-function-calling-6\u001b[0m \u001b[2m               \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[4m0k/tree/26d14ebfe18b1f7b524bd39b404b50af5dc97866?\u001b[0m\u001b[4mrecursive\u001b[0m\u001b[4m=\u001b[0m\u001b[4mfalse\u001b[0m\u001b[4m&\u001b[0m\u001b[4mexpan\u001b[0m \u001b[2m               \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[4md\u001b[0m\u001b[4m=\u001b[0m\u001b[4mfalse\u001b[0m \"HTTP/1.1 200 OK\"                                              \u001b[2m               \u001b[0m\n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/html": [
+       "
                    INFO     HTTP Request: HEAD                                                     _client.py:1025\n",
+       "                             https://huggingface.co/datasets/Salesforce/xlam-function-calling-60k/r                \n",
+       "                             esolve/26d14ebfe18b1f7b524bd39b404b50af5dc97866/dataset_infos.json                    \n",
+       "                             \"HTTP/1.1 404 Not Found\"                                                              \n",
+       "
\n"
+      ],
+      "text/plain": [
+       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m HTTP Request: \u001b[1mHEAD\u001b[0m                                                     \u001b]8;id=8266160;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py\u001b\\\u001b[2m_client.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=8266161;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py#1025\u001b\\\u001b[2m1025\u001b[0m\u001b]8;;\u001b\\\n",
+       "\u001b[2m                    \u001b[0m         \u001b[4mhttps://huggingface.co/datasets/Salesforce/xlam-function-calling-60k/r\u001b[0m \u001b[2m               \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \u001b[4mesolve/26d14ebfe18b1f7b524bd39b404b50af5dc97866/dataset_infos.json\u001b[0m     \u001b[2m               \u001b[0m\n",
+       "\u001b[2m                    \u001b[0m         \"HTTP/1.1 404 Not Found\"                                               \u001b[2m               \u001b[0m\n"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "text/plain": [
+       "{'query': 'Where can I find live giveaways for beta access and games?',\n",
+       " 'tools': '[{\"name\": \"live_giveaways_by_type\", \"description\": \"Retrieve live giveaways from the GamerPower API based on the specified type.\", \"parameters\": {\"type\": {\"description\": \"The type of giveaways to retrieve (e.g., game, loot, beta).\", \"type\": \"str\", \"default\": \"game\"}}}]',\n",
+       " 'answers': '[{\"name\": \"live_giveaways_by_type\", \"arguments\": {\"type\": \"beta\"}}, {\"name\": \"live_giveaways_by_type\", \"arguments\": {\"type\": \"game\"}}]'}"
+      ]
+     },
+     "execution_count": 5,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
    "source": [
-    "import json\n",
     "from datasets import load_dataset\n",
     "\n",
     "raw = load_dataset(\"Salesforce/xlam-function-calling-60k\", split=\"train\")\n",
-    "print(raw)\n",
-    "print(json.dumps({k: raw[0][k] for k in (\"query\", \"tools\", \"answers\")}, indent=2)[:1200])"
+    "{k: raw[0][k] for k in (\"query\", \"tools\", \"answers\")}"
    ]
   },
   {
@@ -167,13 +418,35 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 7,
    "id": "2b17c98c",
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Saving the dataset (1/1 shards): 100%|██████████| 2000/2000 [00:00<00:00, 421665.23 examples/s]\n"
+     ]
+    },
+    {
+     "data": {
+      "text/plain": [
+       "Dataset({\n",
+       "    features: ['answers', 'prompt'],\n",
+       "    num_rows: 2000\n",
+       "})"
+      ]
+     },
+     "execution_count": 7,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
    "source": [
     "import shutil\n",
     "from pathlib import Path\n",
+    "import json\n",
     "\n",
     "N_TRAIN = 2000   # a small subset keeps this demo cheap; the full set is 60k\n",
     "\n",
@@ -222,10 +495,23 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 8,
    "id": "c15e88de",
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "[\n",
+      "  {\n",
+      "    \"role\": \"system\",\n",
+      "    \"content\": \"/no_think\\nYou are an expert in composing function calls. Return exactly one function call that answers the user's request.\\n\\nYou have access to the following tools:\\n\\n[{\\\"name\\\": \\\"user_metadata_information\\\", \\\"description\\\": \\\"Fetch and return metadata information for a specified TikTok user, such as number of followers, followings, avatar URL, description, and more.\\\", \\\"parameters\\\": {\\\"username\\\": {\\\"description\\\": \\\"The TikTok username to fetch metadata for (e.g., \\\\\\\"amazon\\\\\\\").\\\", \\\"type\\\": \\\"str\\\", \\\"default\\\": \\\"tiktok\\\"}, \\\"fresh\\\": {\\\"description\\\": \\\"If set to '1', forces the API to return fresh (non-cached) data. Defaults to '0'.\\\", \\\"type\\\": \\\"str, optional\\\", \\\"default\\\": \\\"0\\\"}}}, {\\\"name\\\": \\\"medias\\\", \\\"description\\\": \\\"Retrieves media posts from a specified Instagram user using the Instagram RapidAPI.\\\", \\\"parameters\\\": {\\\"user_id\\\": {\\\"description\\\": \\\"The ID of the Instagram user whose media posts are to be retrieved.\\\", \\\"type\\\": \\\"str\\\", \\\"default\\\": \\\"25025320\\\"}, \\\"batch_size\\\": {\\\"description\\\": \\\"The number of media posts to retrieve in a single batch, ranging from 1 to 50. Defaults to 20.\\\", \\\"type\\\": \\\"int, optional\\\", \\\"default\\\": \\\"20\\\"}, \\\"next_cursor\\\": {\\\"description\\\": \\\"The cursor for fetching the next set of media posts in pagination. Defaults to None.\\\", \\\"type\\\": \\\"str, optional\\\", \\\"default\\\": \\\"\\\"}}}, {\\\"name\\\": \\\"followers\\\", \\\"description\\\": \\\"Retrieves the list of followers for\n",
+      "\n",
+      "ground truth: [{\"name\": \"followers\", \"arguments\": {\"user_id\": \"17841420039002923\", \"batch_size\": 25}}]\n"
+     ]
+    }
+   ],
    "source": [
     "print(json.dumps(prepared[0][\"prompt\"], indent=2)[:1500])\n",
     "print(\"\\nground truth:\", prepared[0][\"answers\"])"
@@ -241,10 +527,18 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 9,
    "id": "0859c513",
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "s3://sagemaker-us-east-1-754289655784/xlam-grpo/prepared\n"
+     ]
+    }
+   ],
    "source": [
     "train_s3 = sess.upload_data(\"prepared\", bucket=bucket, key_prefix=\"xlam-grpo/prepared\")\n",
     "print(train_s3)"
@@ -267,7 +561,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 10,
    "id": "8dab8b5e",
    "metadata": {},
    "outputs": [],
@@ -278,10 +572,18 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 11,
    "id": "11c358f0",
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Overwriting src/rewards.py\n"
+     ]
+    }
+   ],
    "source": [
     "%%writefile src/rewards.py\n",
     "'''Verifiable rewards for GRPO tool-call training.\n",
@@ -419,10 +721,21 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 12,
    "id": "d91edf68",
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "parsed:      [{'name': 'get_weather', 'arguments': {'city': 'Paris', 'unit': 'c'}}]\n",
+      "exact-match: [1.0]\n",
+      "format:      [1.0]\n",
+      "wrong call:  [0.0]\n"
+     ]
+    }
+   ],
    "source": [
     "from src.rewards import parse_tool_calls, tool_call_reward, format_reward\n",
     "\n",
@@ -449,12 +762,12 @@
     "- `renormalize_logits=True` makes sampling robust if a logits processor creates invalid probabilities.\n",
     "- `attn_implementation=\"sdpa\"` uses the stable attention path for this stack.\n",
     "\n",
-    "Generation uses the Transformers backend.\n"
+    "This notebook leaves TRL's optional vLLM generation path disabled and uses the Transformers generation backend inside `GRPOTrainer`.\n"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 28,
+   "execution_count": 13,
    "id": "f42a218b",
    "metadata": {},
    "outputs": [
@@ -632,15 +945,23 @@
    "id": "b0249124",
    "metadata": {},
    "source": [
-    "ZeRO-3 shards the model, gradients, and optimizer state across the eight A100 GPUs. This run keeps `beta` enabled, so TRL also loads a reference model for the KL term.\n"
+    "DeepSpeed ZeRO-3 shards the model, gradients, and optimizer state across the 8 x A100 40GB GPUs. This run keeps `beta` enabled, so TRL also loads a reference model for the KL term.\n"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 14,
    "id": "395b0169",
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Overwriting src/ds_zero3.json\n"
+     ]
+    }
+   ],
    "source": [
     "%%writefile src/ds_zero3.json\n",
     "{\n",
@@ -668,7 +989,7 @@
    "source": [
     "## Launch the training job\n",
     "\n",
-    "Multi-GPU GRPO needs one process per GPU. `Torchrun` lets the SageMaker SDK start `train.py` directly on all eight A100 GPUs, so the notebook does not need a shell launcher.\n",
+    "Multi-GPU GRPO needs one process per GPU. `Torchrun` lets the SageMaker SDK start `train.py` directly on all 8 x A100 40GB GPUs, so the notebook does not need a shell launcher.\n",
     "\n",
     "This configuration is intentionally small: it is meant to verify that the dataset, reward functions, distributed launch, and logging are wired correctly. Do not expect this short run to materially improve the model; for a real RL training run, increase the number of steps and validate on a held-out set.\n",
     "\n",
@@ -677,5856 +998,10 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 29,
+   "execution_count": null,
    "id": "3204defb",
    "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 10:53:16] INFO     OutputDataConfig compression type not provided. Using default:         defaults.py:165\n",
-       "                             GZIP                                                                                  \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 10:53:16]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m OutputDataConfig compression type not provided. Using default:         \u001b]8;id=17027955;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/train/defaults.py\u001b\\\u001b[2mdefaults.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027956;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/train/defaults.py#165\u001b\\\u001b[2m165\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         GZIP                                                                   \u001b[2m               \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     Training image URI:                                               model_trainer.py:558\n",
-       "                             754289655784.dkr.ecr.us-east-1.amazonaws.com/hf-trl-grpo:sagemake                     \n",
-       "                             r-trl-dev-e63f67e                                                                     \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m Training image URI:                                               \u001b]8;id=17027963;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/train/model_trainer.py\u001b\\\u001b[2mmodel_trainer.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027964;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/train/model_trainer.py#558\u001b\\\u001b[2m558\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m754289655784.\u001b[0mdkr.ecr.us-east-\u001b[1m1.\u001b[0mamazonaws.com/hf-trl-grpo:sagemake \u001b[2m                    \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         r-trl-dev-e63f67e                                                 \u001b[2m                    \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     SageMaker Python SDK will collect telemetry to help us better telemetry_logging.py:110\n",
-       "                             understand our user's needs, diagnose issues, and deliver                             \n",
-       "                             additional features.                                                                  \n",
-       "                             To opt out of telemetry, please disable via TelemetryOptOut                           \n",
-       "                             parameter in SDK defaults config. For more information, refer                         \n",
-       "                             to                                                                                    \n",
-       "                             https://sagemaker.readthedocs.io/en/stable/overview.html#conf                         \n",
-       "                             iguring-and-using-defaults-with-the-sagemaker-python-sdk.                             \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m SageMaker Python SDK will collect telemetry to help us better \u001b]8;id=17027971;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/telemetry/telemetry_logging.py\u001b\\\u001b[2mtelemetry_logging.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027972;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/telemetry/telemetry_logging.py#110\u001b\\\u001b[2m110\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         understand our user's needs, diagnose issues, and deliver     \u001b[2m                        \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         additional features.                                          \u001b[2m                        \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         To opt out of telemetry, please disable via TelemetryOptOut   \u001b[2m                        \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         parameter in SDK defaults config. For more information, refer \u001b[2m                        \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         to                                                            \u001b[2m                        \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[4mhttps://sagemaker.readthedocs.io/en/stable/overview.html#conf\u001b[0m \u001b[2m                        \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[4miguring-and-using-defaults-with-the-sagemaker-python-sdk.\u001b[0m     \u001b[2m                        \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 10:53:22] INFO     Creating training_job resource.                                     resources.py:31116\n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 10:53:22]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m Creating training_job resource.                                     \u001b]8;id=17027979;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027980;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31116\u001b\\\u001b[2m31116\u001b[0m\u001b]8;;\u001b\\\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:01:44] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             Starting training script                                                              \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:01:44]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17027987;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027988;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Starting training script                                            \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             ++ /opt/venv/bin/python3 --version                                                    \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17027995;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17027996;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         ++ /opt/venv/bin/python3 --version                                  \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             Python 3.12.13                                                                        \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028003;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028004;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Python \u001b[1m3.12\u001b[0m.\u001b[1m13\u001b[0m                                                      \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             ++ echo /opt/ml/input/config/resourceconfig.json:                                     \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028011;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028012;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         ++ echo /opt/ml/input/config/resourceconfig.json:                   \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             /opt/ml/input/config/resourceconfig.json:                                             \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028019;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028020;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         /opt/ml/input/config/resourceconfig.json:                           \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             ++ cat /opt/ml/input/config/resourceconfig.json                                       \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028027;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028028;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         ++ cat /opt/ml/input/config/resourceconfig.json                     \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             ++ echo                                                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028035;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028036;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         ++ echo                                                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             {\"current_host\":\"algo-1\",\"current_instance_type\":\"ml.p4d.24xlarge\",                   \n",
-       "                             \"current_group_name\":\"homogeneousCluster\",\"hosts\":[\"algo-1\"],\"insta                   \n",
-       "                             nce_groups\":[{\"instance_group_name\":\"homogeneousCluster\",\"instance_                   \n",
-       "                             type\":\"ml.p4d.24xlarge\",\"hosts\":[\"algo-1\"]}],\"network_interface_nam                   \n",
-       "                             e\":\"eth0\",\"topology\":null}                                                            \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028043;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028044;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m{\u001b[0m\"current_host\":\"algo-1\",\"current_instance_type\":\"ml.p4d.24xlarge\", \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"current_group_name\":\"homogeneousCluster\",\"hosts\":\u001b[1m[\u001b[0m\"algo-1\"\u001b[1m]\u001b[0m,\"insta \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         nce_groups\":\u001b[1m[\u001b[0m\u001b[1m{\u001b[0m\"instance_group_name\":\"homogeneousCluster\",\"instance_ \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         type\":\"ml.p4d.24xlarge\",\"hosts\":\u001b[1m[\u001b[0m\"algo-1\"\u001b[1m]\u001b[0m\u001b[1m}\u001b[0m\u001b[1m]\u001b[0m,\"network_interface_nam \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         e\":\"eth0\",\"topology\":null\u001b[1m}\u001b[0m                                          \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             /opt/ml/input/config/inputdataconfig.json:                                            \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028051;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028052;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         /opt/ml/input/config/inputdataconfig.json:                          \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             ++ echo /opt/ml/input/config/inputdataconfig.json:                                    \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028059;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028060;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         ++ echo /opt/ml/input/config/inputdataconfig.json:                  \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             ++ cat /opt/ml/input/config/inputdataconfig.json                                      \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028067;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028068;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         ++ cat /opt/ml/input/config/inputdataconfig.json                    \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             ++ echo                                                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028075;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028076;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         ++ echo                                                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             ++ echo 'Setting up environment variables'                                            \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028083;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028084;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         ++ echo 'Setting up environment variables'                          \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             ++ /opt/venv/bin/python3                                                              \n",
-       "                             /opt/ml/input/data/sm_drivers/scripts/environment.py                                  \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028091;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028092;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         ++ /opt/venv/bin/python3                                            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         /opt/ml/input/data/sm_drivers/scripts/environment.py                \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             {\"code\":{\"TrainingInputMode\":\"File\",\"S3DistributionType\":\"FullyRepl                   \n",
-       "                             icated\",\"RecordWrapperType\":\"None\"},\"sm_drivers\":{\"TrainingInputMod                   \n",
-       "                             e\":\"File\",\"S3DistributionType\":\"FullyReplicated\",\"RecordWrapperType                   \n",
-       "                             \":\"None\"},\"train\":{\"TrainingInputMode\":\"File\",\"S3DistributionType\":                   \n",
-       "                             \"FullyReplicated\",\"RecordWrapperType\":\"None\"}}                                        \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028099;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028100;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m{\u001b[0m\"code\":\u001b[1m{\u001b[0m\"TrainingInputMode\":\"File\",\"S3DistributionType\":\"FullyRepl \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         icated\",\"RecordWrapperType\":\"None\"\u001b[1m}\u001b[0m,\"sm_drivers\":\u001b[1m{\u001b[0m\"TrainingInputMod \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         e\":\"File\",\"S3DistributionType\":\"FullyReplicated\",\"RecordWrapperType \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \":\"None\"\u001b[1m}\u001b[0m,\"train\":\u001b[1m{\u001b[0m\"TrainingInputMode\":\"File\",\"S3DistributionType\": \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"FullyReplicated\",\"RecordWrapperType\":\"None\"\u001b[1m}\u001b[0m\u001b[1m}\u001b[0m                      \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             Setting up environment variables                                                      \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028107;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028108;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Setting up environment variables                                    \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             No Neurons detected (normal if no neurons installed)                                  \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028115;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028116;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         No Neurons detected \u001b[1m(\u001b[0mnormal if no neurons installed\u001b[1m)\u001b[0m                \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             Environment Variables:                                                                \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028123;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028124;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Environment Variables:                                              \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             NV_LIBCUBLAS_VERSION=13.1.0.3-1                                                       \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028131;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028132;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         NV_LIBCUBLAS_VERSION=13.1.0.3-\u001b[1m1\u001b[0m                                     \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             NVIDIA_VISIBLE_DEVICES=all                                                            \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028139;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028140;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         NVIDIA_VISIBLE_DEVICES=all                                          \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             PYTHONUNBUFFERED=1                                                                    \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028147;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028148;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         PYTHONUNBUFFERED=\u001b[1m1\u001b[0m                                                  \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             AWS_CONTAINER_CREDENTIALS_RELATIVE_URI=******                                         \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028155;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028156;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         AWS_CONTAINER_CREDENTIALS_RELATIVE_URI=******                       \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SAGEMAKER_TRAINING_MODULE=sagemaker_pytorch_container.training:main                   \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028163;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028164;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SAGEMAKER_TRAINING_MODULE=sagemaker_pytorch_container.training:main \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             HOSTNAME=ip-10-0-71-109.ec2.internal                                                  \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028171;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028172;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         HOSTNAME=ip-\u001b[1m10\u001b[0m-\u001b[1m0\u001b[0m-\u001b[1m71\u001b[0m-\u001b[1m109.\u001b[0mec2.internal                                \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             NVIDIA_REQUIRE_CUDA=cuda>=13.0 brand=unknown,driver>=535,driver<536                   \n",
-       "                             brand=grid,driver>=535,driver<536                                                     \n",
-       "                             brand=tesla,driver>=535,driver<536                                                    \n",
-       "                             brand=nvidia,driver>=535,driver<536                                                   \n",
-       "                             brand=quadro,driver>=535,driver<536                                                   \n",
-       "                             brand=quadrortx,driver>=535,driver<536                                                \n",
-       "                             brand=nvidiartx,driver>=535,driver<536                                                \n",
-       "                             brand=vapps,driver>=535,driver<536 brand=vpc,driver>=535,driver<536                   \n",
-       "                             brand=vcs,driver>=535,driver<536 brand=vws,driver>=535,driver<536                     \n",
-       "                             brand=cloudgaming,driver>=535,driver<536                                              \n",
-       "                             brand=unknown,driver>=550,driver<551                                                  \n",
-       "                             brand=grid,driver>=550,driver<551                                                     \n",
-       "                             brand=tesla,driver>=550,driver<551                                                    \n",
-       "                             brand=nvidia,driver>=550,driver<551                                                   \n",
-       "                             brand=quadro,driver>=550,driver<551                                                   \n",
-       "                             brand=quadrortx,driver>=550,driver<551                                                \n",
-       "                             brand=nvidiartx,driver>=550,driver<551                                                \n",
-       "                             brand=vapps,driver>=550,driver<551 brand=vpc,driver>=550,driver<551                   \n",
-       "                             brand=vcs,driver>=550,driver<551 brand=vws,driver>=550,driver<551                     \n",
-       "                             brand=cloudgaming,driver>=550,driver<551                                              \n",
-       "                             brand=unknown,driver>=565,driver<566                                                  \n",
-       "                             brand=grid,driver>=565,driver<566                                                     \n",
-       "                             brand=tesla,driver>=565,driver<566                                                    \n",
-       "                             brand=nvidia,driver>=565,driver<566                                                   \n",
-       "                             brand=quadro,driver>=565,driver<566                                                   \n",
-       "                             brand=quadrortx,driver>=565,driver<566                                                \n",
-       "                             brand=nvidiartx,driver>=565,driver<566                                                \n",
-       "                             brand=vapps,driver>=565,driver<566 brand=vpc,driver>=565,driver<566                   \n",
-       "                             brand=vcs,driver>=565,driver<566 brand=vws,driver>=565,driver<566                     \n",
-       "                             brand=cloudgaming,driver>=565,driver<566                                              \n",
-       "                             brand=unknown,driver>=570,driver<571                                                  \n",
-       "                             brand=grid,driver>=570,driver<571                                                     \n",
-       "                             brand=tesla,driver>=570,driver<571                                                    \n",
-       "                             brand=nvidia,driver>=570,driver<571                                                   \n",
-       "                             brand=quadro,driver>=570,driver<571                                                   \n",
-       "                             brand=quadrortx,driver>=570,driver<571                                                \n",
-       "                             brand=nvidiartx,driver>=570,driver<571                                                \n",
-       "                             brand=vapps,driver>=570,driver<571 brand=vpc,driver>=570,driver<571                   \n",
-       "                             brand=vcs,driver>=570,driver<571 brand=vws,driver>=570,driver<571                     \n",
-       "                             brand=cloudgaming,driver>=570,driver<571                                              \n",
-       "                             brand=unknown,driver>=575,driver<576                                                  \n",
-       "                             brand=grid,driver>=575,driver<576                                                     \n",
-       "                             brand=tesla,driver>=575,driver<576                                                    \n",
-       "                             brand=nvidia,driver>=575,driver<576                                                   \n",
-       "                             brand=quadro,driver>=575,driver<576                                                   \n",
-       "                             brand=quadrortx,driver>=575,driver<576                                                \n",
-       "                             brand=nvidiartx,driver>=575,driver<576                                                \n",
-       "                             brand=vapps,driver>=575,driver<576 brand=vpc,driver>=575,driver<576                   \n",
-       "                             brand=vcs,driver>=575,driver<576 brand=vws,driver>=575,driver<576                     \n",
-       "                             brand=cloudgaming,driver>=575,driver<576                                              \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028179;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028180;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         NVIDIA_REQUIRE_CUDA=cuda>=\u001b[1m13.0\u001b[0m brand=unknown,driver>=\u001b[1m535\u001b[0m,driver\u001b[1m<\u001b[0m\u001b[1m536\u001b[0m \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=grid,driver>=\u001b[1m535\u001b[0m,driver<\u001b[1m536\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=tesla,driver>=\u001b[1m535\u001b[0m,driver<\u001b[1m536\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=nvidia,driver>=\u001b[1m535\u001b[0m,driver<\u001b[1m536\u001b[0m                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=quadro,driver>=\u001b[1m535\u001b[0m,driver<\u001b[1m536\u001b[0m                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=quadrortx,driver>=\u001b[1m535\u001b[0m,driver<\u001b[1m536\u001b[0m                              \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=nvidiartx,driver>=\u001b[1m535\u001b[0m,driver<\u001b[1m536\u001b[0m                              \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=vapps,driver>=\u001b[1m535\u001b[0m,driver<\u001b[1m536\u001b[0m brand=vpc,driver>=\u001b[1m535\u001b[0m,driver<\u001b[1m536\u001b[0m \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=vcs,driver>=\u001b[1m535\u001b[0m,driver<\u001b[1m536\u001b[0m brand=vws,driver>=\u001b[1m535\u001b[0m,driver<\u001b[1m536\u001b[0m   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=cloudgaming,driver>=\u001b[1m535\u001b[0m,driver<\u001b[1m536\u001b[0m                            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=unknown,driver>=\u001b[1m550\u001b[0m,driver<\u001b[1m551\u001b[0m                                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=grid,driver>=\u001b[1m550\u001b[0m,driver<\u001b[1m551\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=tesla,driver>=\u001b[1m550\u001b[0m,driver<\u001b[1m551\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=nvidia,driver>=\u001b[1m550\u001b[0m,driver<\u001b[1m551\u001b[0m                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=quadro,driver>=\u001b[1m550\u001b[0m,driver<\u001b[1m551\u001b[0m                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=quadrortx,driver>=\u001b[1m550\u001b[0m,driver<\u001b[1m551\u001b[0m                              \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=nvidiartx,driver>=\u001b[1m550\u001b[0m,driver<\u001b[1m551\u001b[0m                              \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=vapps,driver>=\u001b[1m550\u001b[0m,driver<\u001b[1m551\u001b[0m brand=vpc,driver>=\u001b[1m550\u001b[0m,driver<\u001b[1m551\u001b[0m \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=vcs,driver>=\u001b[1m550\u001b[0m,driver<\u001b[1m551\u001b[0m brand=vws,driver>=\u001b[1m550\u001b[0m,driver<\u001b[1m551\u001b[0m   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=cloudgaming,driver>=\u001b[1m550\u001b[0m,driver<\u001b[1m551\u001b[0m                            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=unknown,driver>=\u001b[1m565\u001b[0m,driver<\u001b[1m566\u001b[0m                                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=grid,driver>=\u001b[1m565\u001b[0m,driver<\u001b[1m566\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=tesla,driver>=\u001b[1m565\u001b[0m,driver<\u001b[1m566\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=nvidia,driver>=\u001b[1m565\u001b[0m,driver<\u001b[1m566\u001b[0m                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=quadro,driver>=\u001b[1m565\u001b[0m,driver<\u001b[1m566\u001b[0m                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=quadrortx,driver>=\u001b[1m565\u001b[0m,driver<\u001b[1m566\u001b[0m                              \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=nvidiartx,driver>=\u001b[1m565\u001b[0m,driver<\u001b[1m566\u001b[0m                              \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=vapps,driver>=\u001b[1m565\u001b[0m,driver<\u001b[1m566\u001b[0m brand=vpc,driver>=\u001b[1m565\u001b[0m,driver<\u001b[1m566\u001b[0m \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=vcs,driver>=\u001b[1m565\u001b[0m,driver<\u001b[1m566\u001b[0m brand=vws,driver>=\u001b[1m565\u001b[0m,driver<\u001b[1m566\u001b[0m   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=cloudgaming,driver>=\u001b[1m565\u001b[0m,driver<\u001b[1m566\u001b[0m                            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=unknown,driver>=\u001b[1m570\u001b[0m,driver<\u001b[1m571\u001b[0m                                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=grid,driver>=\u001b[1m570\u001b[0m,driver<\u001b[1m571\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=tesla,driver>=\u001b[1m570\u001b[0m,driver<\u001b[1m571\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=nvidia,driver>=\u001b[1m570\u001b[0m,driver<\u001b[1m571\u001b[0m                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=quadro,driver>=\u001b[1m570\u001b[0m,driver<\u001b[1m571\u001b[0m                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=quadrortx,driver>=\u001b[1m570\u001b[0m,driver<\u001b[1m571\u001b[0m                              \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=nvidiartx,driver>=\u001b[1m570\u001b[0m,driver<\u001b[1m571\u001b[0m                              \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=vapps,driver>=\u001b[1m570\u001b[0m,driver<\u001b[1m571\u001b[0m brand=vpc,driver>=\u001b[1m570\u001b[0m,driver<\u001b[1m571\u001b[0m \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=vcs,driver>=\u001b[1m570\u001b[0m,driver<\u001b[1m571\u001b[0m brand=vws,driver>=\u001b[1m570\u001b[0m,driver<\u001b[1m571\u001b[0m   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=cloudgaming,driver>=\u001b[1m570\u001b[0m,driver<\u001b[1m571\u001b[0m                            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=unknown,driver>=\u001b[1m575\u001b[0m,driver<\u001b[1m576\u001b[0m                                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=grid,driver>=\u001b[1m575\u001b[0m,driver<\u001b[1m576\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=tesla,driver>=\u001b[1m575\u001b[0m,driver<\u001b[1m576\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=nvidia,driver>=\u001b[1m575\u001b[0m,driver<\u001b[1m576\u001b[0m                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=quadro,driver>=\u001b[1m575\u001b[0m,driver<\u001b[1m576\u001b[0m                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=quadrortx,driver>=\u001b[1m575\u001b[0m,driver<\u001b[1m576\u001b[0m                              \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=nvidiartx,driver>=\u001b[1m575\u001b[0m,driver<\u001b[1m576\u001b[0m                              \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=vapps,driver>=\u001b[1m575\u001b[0m,driver<\u001b[1m576\u001b[0m brand=vpc,driver>=\u001b[1m575\u001b[0m,driver<\u001b[1m576\u001b[0m \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=vcs,driver>=\u001b[1m575\u001b[0m,driver<\u001b[1m576\u001b[0m brand=vws,driver>=\u001b[1m575\u001b[0m,driver<\u001b[1m576\u001b[0m   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         brand=cloudgaming,driver\u001b[1m>\u001b[0m=\u001b[1m575\u001b[0m,driver<\u001b[1m576\u001b[0m                            \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             NV_NVTX_VERSION=13.0.85-1                                                             \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028187;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028188;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         NV_NVTX_VERSION=\u001b[1m13\u001b[0m\u001b[1m.0\u001b[0m.\u001b[1m85\u001b[0m-\u001b[1m1\u001b[0m                                           \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             NV_LIBNPP_VERSION=13.0.1.2-1                                                          \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028195;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028196;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         NV_LIBNPP_VERSION=13.0.1.2-\u001b[1m1\u001b[0m                                        \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             AWS_REGION=us-east-1                                                                  \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028203;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028204;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         AWS_REGION=us-east-\u001b[1m1\u001b[0m                                                \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             PWD=/workspace                                                                        \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028211;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028212;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         PWD=/workspace                                                      \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SAGEMAKER_MANAGED_WARMPOOL_CACHE_DIRECTORY=/opt/ml/sagemaker/warmpo                   \n",
-       "                             olcache                                                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028219;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028220;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SAGEMAKER_MANAGED_WARMPOOL_CACHE_DIRECTORY=/opt/ml/sagemaker/warmpo \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         olcache                                                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             NVIDIA_DRIVER_CAPABILITIES=compute,utility,compat32,graphics,video                    \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028227;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028228;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         NVIDIA_DRIVER_CAPABILITIES=compute,utility,compat32,graphics,video  \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             NV_LIBNPP_PACKAGE=libnpp-13-0-13.0.1.2-1                                              \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028235;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028236;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         NV_LIBNPP_PACKAGE=libnpp-\u001b[1m13\u001b[0m-\u001b[1m0\u001b[0m-13.0.1.2-\u001b[1m1\u001b[0m                            \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             NCCL_DEBUG=WARN                                                                       \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028243;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028244;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         NCCL_DEBUG=WARN                                                     \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             NVIDIA_PRODUCT_NAME=CUDA                                                              \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028251;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028252;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         NVIDIA_PRODUCT_NAME=CUDA                                            \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             NV_CUDA_CUDART_VERSION=13.0.96-1                                                      \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028259;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028260;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         NV_CUDA_CUDART_VERSION=\u001b[1m13\u001b[0m\u001b[1m.0\u001b[0m.\u001b[1m96\u001b[0m-\u001b[1m1\u001b[0m                                    \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             HOME=/root                                                                            \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028267;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028268;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         HOME=/root                                                          \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             LANG=C.UTF-8                                                                          \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028275;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028276;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         LANG=C.UTF-\u001b[1m8\u001b[0m                                                        \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             CUDA_VERSION=13.0.2                                                                   \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028283;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028284;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         CUDA_VERSION=\u001b[1m13\u001b[0m\u001b[1m.0\u001b[0m.\u001b[1m2\u001b[0m                                                 \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             HF_HUB_USER_AGENT_ORIGIN=aws:sagemaker:gpu-cuda:training:hf-pytorch                   \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028291;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028292;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         HF_HUB_USER_AGENT_ORIGIN=aws:sagemaker:gpu-cuda:training:hf-pytorch \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             DMLC_INTERFACE=eth0                                                                   \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028299;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028300;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         DMLC_INTERFACE=eth0                                                 \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             HF_TOKEN=******                                                                       \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028307;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028308;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         HF_TOKEN=******                                                     \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             PKG_CMD=yum                                                                           \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028315;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028316;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         PKG_CMD=yum                                                         \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             PYTHONIOENCODING=UTF-8                                                                \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028323;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028324;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         PYTHONIOENCODING=UTF-\u001b[1m8\u001b[0m                                              \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SHLVL=1                                                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028331;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028332;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SHLVL=\u001b[1m1\u001b[0m                                                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             NV_CUDA_LIB_VERSION=13.0.2-1                                                          \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028339;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028340;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         NV_CUDA_LIB_VERSION=\u001b[1m13\u001b[0m\u001b[1m.0\u001b[0m.\u001b[1m2\u001b[0m-\u001b[1m1\u001b[0m                                        \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             NVARCH=x86_64                                                                         \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028347;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028348;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         NVARCH=x86_64                                                       \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             PYTHONDONTWRITEBYTECODE=1                                                             \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028355;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028356;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         PYTHONDONTWRITEBYTECODE=\u001b[1m1\u001b[0m                                           \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             LD_LIBRARY_PATH=/opt/venv/lib/python3.12/site-packages/nvidia/cudnn                   \n",
-       "                             /lib:/opt/venv/lib/python3.12/site-packages/nvidia/nccl/lib:/opt/am                   \n",
-       "                             azon/ofi-nccl/lib64:/opt/amazon/openmpi/lib:/opt/amazon/openmpi/lib                   \n",
-       "                             64:/opt/amazon/efa/lib:/opt/amazon/efa/lib64:/usr/local/cuda/lib64:                   \n",
-       "                             /usr/local/lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/l                   \n",
-       "                             ocal/cuda/lib64                                                                       \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028363;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028364;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         LD_LIBRARY_PATH=/opt/venv/lib/python3.12/site-packages/nvidia/cudnn \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         /lib:/opt/venv/lib/python3.12/site-packages/nvidia/nccl/lib:/opt/am \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         azon/ofi-nccl/lib64:/opt/amazon/openmpi/lib:/opt/amazon/openmpi/lib \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         64:/opt/amazon/efa/lib:/opt/amazon/efa/lib64:/usr/local/cuda/lib64: \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         /usr/local/lib:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/l \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         ocal/cuda/lib64                                                     \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             TRAINING_JOB_NAME=smolm3-grpo-toolcall-20260626-085316-202606261053                   \n",
-       "                             16                                                                                    \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028371;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028372;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         TRAINING_JOB_NAME=smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m202606261053\u001b[0m \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m16\u001b[0m                                                                  \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             LC_ALL=C.UTF-8                                                                        \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028379;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028380;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         LC_ALL=C.UTF-\u001b[1m8\u001b[0m                                                      \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             TRAINING_JOB_ARN=arn:aws:sagemaker:us-east-1:754289655784:training-                   \n",
-       "                             job/smolm3-grpo-toolcall-20260626-085316-20260626105316                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028387;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028388;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         TRAINING_JOB_ARN=arn:aws:sagemaker:us-east-1:754289655784:training- \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         job/smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             CUDA_HOME=/usr/local/cuda                                                             \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028395;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028396;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         CUDA_HOME=/usr/local/cuda                                           \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             PATH=/opt/venv/bin:/opt/amazon/openmpi/bin:/opt/amazon/efa/bin:/usr                   \n",
-       "                             /local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/loca                   \n",
-       "                             l/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin                                   \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028403;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028404;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         PATH=/opt/venv/bin:/opt/amazon/openmpi/bin:/opt/amazon/efa/bin:/usr \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         /local/cuda/bin:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/loca \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         l/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin                 \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             UV_PROJECT_ENVIRONMENT=/opt/venv                                                      \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028411;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028412;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         UV_PROJECT_ENVIRONMENT=/opt/venv                                    \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             DLC_CONTAINER_TYPE=training                                                           \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028419;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028420;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         DLC_CONTAINER_TYPE=training                                         \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             _=/opt/venv/bin/python3                                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028427;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028428;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         _=/opt/venv/bin/python3                                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:01:45] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_MODEL_DIR=/opt/ml/model                                                            \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:01:45]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028435;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028436;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_MODEL_DIR=/opt/ml/model                                          \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_INPUT_DIR=/opt/ml/input                                                            \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028443;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028444;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_IN\u001b[1mPUT\u001b[0m_DIR=/opt/ml/input                                          \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_INPUT_DATA_DIR=/opt/ml/input/data                                                  \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028451;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028452;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_IN\u001b[1mPUT\u001b[0m_DATA_DIR=/opt/ml/input/data                                \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_INPUT_CONFIG_DIR=/opt/ml/input/config                                              \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028459;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028460;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_IN\u001b[1mPUT\u001b[0m_CONFIG_DIR=/opt/ml/input/config                            \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_OUTPUT_DIR=/opt/ml/output                                                          \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028467;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028468;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_OUT\u001b[1mPUT\u001b[0m_DIR=/opt/ml/output                                        \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_OUTPUT_FAILURE=/opt/ml/output/failure                                              \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028475;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028476;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_OUT\u001b[1mPUT\u001b[0m_FAILURE=/opt/ml/output/failure                            \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_OUTPUT_DATA_DIR=/opt/ml/output/data                                                \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028483;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028484;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_OUT\u001b[1mPUT\u001b[0m_DATA_DIR=/opt/ml/output/data                              \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_LOG_LEVEL=20                                                                       \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028491;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028492;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_LOG_LEVEL=\u001b[1m20\u001b[0m                                                     \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_MASTER_ADDR=algo-1                                                                 \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028499;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028500;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_MASTER_ADDR=algo-\u001b[1m1\u001b[0m                                               \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_MASTER_PORT=7777                                                                   \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028507;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028508;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_MASTER_PORT=\u001b[1m7777\u001b[0m                                                 \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_SOURCE_DIR=/opt/ml/input/data/code                                                 \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028515;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028516;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_SOURCE_DIR=/opt/ml/input/data/code                               \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_ENTRY_SCRIPT=train.py                                                              \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028523;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028524;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_ENTRY_SCRIPT=train.py                                            \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_DISTRIBUTED_DRIVER_DIR=/opt/ml/input/data/sm_drivers/distributed                   \n",
-       "                             _drivers                                                                              \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028531;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028532;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_DISTRIBUTED_DRIVER_DIR=/opt/ml/input/data/sm_drivers/distributed \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         _drivers                                                            \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_DISTRIBUTED_CONFIG={\"process_count_per_node\": 8, \"smp\": null}                      \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028539;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028540;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_DISTRIBUTED_CONFIG=\u001b[1m{\u001b[0m\"process_count_per_node\": \u001b[1m8\u001b[0m, \"smp\": null\u001b[1m}\u001b[0m    \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_CHANNEL_CODE=/opt/ml/input/data/code                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028547;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028548;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_CHANNEL_CODE=/opt/ml/input/data/code                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_CHANNEL_SM_DRIVERS=/opt/ml/input/data/sm_drivers                                   \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028555;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028556;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_CHANNEL_SM_DRIVERS=/opt/ml/input/data/sm_drivers                 \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_CHANNEL_TRAIN=/opt/ml/input/data/train                                             \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028563;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028564;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_CHANNEL_TRAIN=/opt/ml/input/data/train                           \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_CHANNELS=['code', 'sm_drivers', 'train']                                           \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028571;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028572;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_CHANNELS=\u001b[1m[\u001b[0m'code', 'sm_drivers', 'train'\u001b[1m]\u001b[0m                         \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_HP_BETA=0.1                                                                        \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028579;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028580;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_HP_BETA=\u001b[1m0\u001b[0m\u001b[1m.1\u001b[0m                                                      \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_HP_DEEPSPEED=ds_zero3.json                                                         \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028587;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028588;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_HP_DEEPSPEED=ds_zero3.json                                       \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_HP_GRADIENT_ACCUMULATION_STEPS=8                                                   \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028595;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028596;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_HP_GRADIENT_ACCUMULATION_STEPS=\u001b[1m8\u001b[0m                                 \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_HP_GRADIENT_CHECKPOINTING=True                                                     \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028603;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028604;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_HP_GRADIENT_CHECKPOINTING=\u001b[3mTrue\u001b[0m                                   \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_HP_LEARNING_RATE=1e-08                                                             \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028611;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028612;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_HP_LEARNING_RATE=\u001b[1m1e\u001b[0m\u001b[1m-08\u001b[0m                                           \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_HP_LOG_COMPLETIONS=False                                                           \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028619;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028620;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_HP_LOG_COMPLETIONS=\u001b[3mFalse\u001b[0m                                         \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_HP_MAX_COMPLETION_LENGTH=128                                                       \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028627;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028628;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_HP_MAX_COMPLETION_LENGTH=\u001b[1m128\u001b[0m                                     \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_HP_MAX_GRAD_NORM=0.05                                                              \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028635;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028636;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_HP_MAX_GRAD_NORM=\u001b[1m0\u001b[0m\u001b[1m.05\u001b[0m                                            \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_HP_MAX_STEPS=50                                                                    \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028643;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028644;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_HP_MAX_STEPS=\u001b[1m50\u001b[0m                                                  \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_HP_MODEL_ID=HuggingFaceTB/SmolLM3-3B                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028651;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028652;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_HP_MODEL_ID=HuggingFaceTB/SmolLM3-3B                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_HP_NUM_COMPLETIONS_TO_PRINT=8                                                      \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028659;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028660;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_HP_NUM_COMPLETIONS_TO_PRINT=\u001b[1m8\u001b[0m                                    \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_HP_NUM_GENERATIONS=8                                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028667;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028668;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_HP_NUM_GENERATIONS=\u001b[1m8\u001b[0m                                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_HP_PER_DEVICE_TRAIN_BATCH_SIZE=2                                                   \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028675;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028676;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_HP_PER_DEVICE_TRAIN_BATCH_SIZE=\u001b[1m2\u001b[0m                                 \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_HP_REMOVE_INVALID_VALUES=False                                                     \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028683;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028684;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_HP_REMOVE_INVALID_VALUES=\u001b[3mFalse\u001b[0m                                   \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_HP_RENORMALIZE_LOGITS=True                                                         \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028691;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028692;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_HP_RENORMALIZE_LOGITS=\u001b[3mTrue\u001b[0m                                       \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_HP_REPETITION_PENALTY=1.05                                                         \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028699;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028700;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_HP_REPETITION_PENALTY=\u001b[1m1\u001b[0m\u001b[1m.05\u001b[0m                                       \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_HP_REWARD_WEIGHTS=1.0,0.5                                                          \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028707;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028708;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_HP_REWARD_WEIGHTS=\u001b[1m1\u001b[0m\u001b[1m.0\u001b[0m,\u001b[1m0.5\u001b[0m                                        \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_HP_STOP_ON_COLLAPSE=True                                                           \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028715;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028716;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_HP_STOP_ON_COLLAPSE=\u001b[3mTrue\u001b[0m                                         \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_HP_TEMPERATURE=0.7                                                                 \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028723;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028724;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_HP_TEMPERATURE=\u001b[1m0\u001b[0m\u001b[1m.7\u001b[0m                                               \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_HP_TOP_K=0                                                                         \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028731;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028732;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_HP_TOP_K=\u001b[1m0\u001b[0m                                                       \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_HP_TOP_P=1.0                                                                       \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028739;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028740;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_HP_TOP_P=\u001b[1m1\u001b[0m\u001b[1m.0\u001b[0m                                                     \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_HPS={\"beta\": 0.1, \"deepspeed\": \"ds_zero3.json\",                                    \n",
-       "                             \"gradient_accumulation_steps\": 8, \"gradient_checkpointing\": true,                     \n",
-       "                             \"learning_rate\": 1e-08, \"log_completions\": false,                                     \n",
-       "                             \"max_completion_length\": 128, \"max_grad_norm\": 0.05, \"max_steps\":                     \n",
-       "                             50, \"model_id\": \"HuggingFaceTB/SmolLM3-3B\",                                           \n",
-       "                             \"num_completions_to_print\": 8, \"num_generations\": 8,                                  \n",
-       "                             \"per_device_train_batch_size\": 2, \"remove_invalid_values\": false,                     \n",
-       "                             \"renormalize_logits\": true, \"repetition_penalty\": 1.05,                               \n",
-       "                             \"reward_weights\": \"1.0,0.5\", \"stop_on_collapse\": true,                                \n",
-       "                             \"temperature\": 0.7, \"top_k\": 0, \"top_p\": 1.0}                                         \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028747;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028748;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_HPS=\u001b[1m{\u001b[0m\"beta\": \u001b[1m0.1\u001b[0m, \"deepspeed\": \"ds_zero3.json\",                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"gradient_accumulation_steps\": \u001b[1m8\u001b[0m, \"gradient_checkpointing\": true,   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"learning_rate\": \u001b[1m1e-08\u001b[0m, \"log_completions\": false,                   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"max_completion_length\": \u001b[1m128\u001b[0m, \"max_grad_norm\": \u001b[1m0.05\u001b[0m, \"max_steps\":   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m50\u001b[0m, \"model_id\": \"HuggingFaceTB/SmolLM3-3B\",                         \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"num_completions_to_print\": \u001b[1m8\u001b[0m, \"num_generations\": \u001b[1m8\u001b[0m,                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"per_device_train_batch_size\": \u001b[1m2\u001b[0m, \"remove_invalid_values\": false,   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"renormalize_logits\": true, \"repetition_penalty\": \u001b[1m1.05\u001b[0m,             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"reward_weights\": \"1.0,0.5\", \"stop_on_collapse\": true,              \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"temperature\": \u001b[1m0.7\u001b[0m, \"top_k\": \u001b[1m0\u001b[0m, \"top_p\": \u001b[1m1.0\u001b[0m\u001b[1m}\u001b[0m                       \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_CURRENT_HOST=algo-1                                                                \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028755;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028756;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_CURRENT_HOST=algo-\u001b[1m1\u001b[0m                                              \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_CURRENT_INSTANCE_TYPE=ml.p4d.24xlarge                                              \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028763;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028764;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_CURRENT_INSTANCE_TYPE=ml.p4d.24xlarge                            \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_HOSTS=['algo-1']                                                                   \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028771;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028772;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_HOSTS=\u001b[1m[\u001b[0m'algo-1'\u001b[1m]\u001b[0m                                                 \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_NETWORK_INTERFACE_NAME=eth0                                                        \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028779;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028780;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_NETWORK_INTERFACE_NAME=eth0                                      \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_HOST_COUNT=1                                                                       \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028787;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028788;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_HOST_COUNT=\u001b[1m1\u001b[0m                                                     \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_CURRENT_HOST_RANK=0                                                                \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028795;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028796;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_CURRENT_HOST_RANK=\u001b[1m0\u001b[0m                                              \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_NUM_CPUS=96                                                                        \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028803;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028804;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_NUM_CPUS=\u001b[1m96\u001b[0m                                                      \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_NUM_GPUS=8                                                                         \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028811;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028812;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_NUM_GPUS=\u001b[1m8\u001b[0m                                                       \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_NUM_NEURONS=0                                                                      \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028819;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028820;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_NUM_NEURONS=\u001b[1m0\u001b[0m                                                    \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_RESOURCE_CONFIG={\"current_host\": \"algo-1\",                                         \n",
-       "                             \"current_instance_type\": \"ml.p4d.24xlarge\", \"current_group_name\":                     \n",
-       "                             \"homogeneousCluster\", \"hosts\": [\"algo-1\"], \"instance_groups\":                         \n",
-       "                             [{\"instance_group_name\": \"homogeneousCluster\", \"instance_type\":                       \n",
-       "                             \"ml.p4d.24xlarge\", \"hosts\": [\"algo-1\"]}], \"network_interface_name\":                   \n",
-       "                             \"eth0\", \"topology\": null}                                                             \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028827;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028828;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_RESOURCE_CONFIG=\u001b[1m{\u001b[0m\"current_host\": \"algo-1\",                       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"current_instance_type\": \"ml.p4d.24xlarge\", \"current_group_name\":   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"homogeneousCluster\", \"hosts\": \u001b[1m[\u001b[0m\"algo-1\"\u001b[1m]\u001b[0m, \"instance_groups\":       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m[\u001b[0m\u001b[1m{\u001b[0m\"instance_group_name\": \"homogeneousCluster\", \"instance_type\":     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"ml.p4d.24xlarge\", \"hosts\": \u001b[1m[\u001b[0m\"algo-1\"\u001b[1m]\u001b[0m\u001b[1m}\u001b[0m\u001b[1m]\u001b[0m, \"network_interface_name\": \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"eth0\", \"topology\": null\u001b[1m}\u001b[0m                                           \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_INPUT_DATA_CONFIG={\"code\": {\"TrainingInputMode\": \"File\",                           \n",
-       "                             \"S3DistributionType\": \"FullyReplicated\", \"RecordWrapperType\":                         \n",
-       "                             \"None\"}, \"sm_drivers\": {\"TrainingInputMode\": \"File\",                                  \n",
-       "                             \"S3DistributionType\": \"FullyReplicated\", \"RecordWrapperType\":                         \n",
-       "                             \"None\"}, \"train\": {\"TrainingInputMode\": \"File\",                                       \n",
-       "                             \"S3DistributionType\": \"FullyReplicated\", \"RecordWrapperType\":                         \n",
-       "                             \"None\"}}                                                                              \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028835;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028836;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_IN\u001b[1mPUT\u001b[0m_DATA_CONFIG=\u001b[1m{\u001b[0m\"code\": \u001b[1m{\u001b[0m\"TrainingInputMode\": \"File\",         \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"S3DistributionType\": \"FullyReplicated\", \"RecordWrapperType\":       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"None\"\u001b[1m}\u001b[0m, \"sm_drivers\": \u001b[1m{\u001b[0m\"TrainingInputMode\": \"File\",                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"S3DistributionType\": \"FullyReplicated\", \"RecordWrapperType\":       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"None\"\u001b[1m}\u001b[0m, \"train\": \u001b[1m{\u001b[0m\"TrainingInputMode\": \"File\",                     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"S3DistributionType\": \"FullyReplicated\", \"RecordWrapperType\":       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"None\"\u001b[1m}\u001b[0m\u001b[1m}\u001b[0m                                                            \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             SM_TRAINING_ENV={\"channel_input_dirs\": {\"code\":                                       \n",
-       "                             \"/opt/ml/input/data/code\", \"sm_drivers\":                                              \n",
-       "                             \"/opt/ml/input/data/sm_drivers\", \"train\":                                             \n",
-       "                             \"/opt/ml/input/data/train\"}, \"current_host\": \"algo-1\",                                \n",
-       "                             \"current_instance_type\": \"ml.p4d.24xlarge\", \"hosts\": [\"algo-1\"],                      \n",
-       "                             \"master_addr\": \"algo-1\", \"master_port\": 7777, \"hyperparameters\":                      \n",
-       "                             {\"beta\": 0.1, \"deepspeed\": \"ds_zero3.json\",                                           \n",
-       "                             \"gradient_accumulation_steps\": 8, \"gradient_checkpointing\": true,                     \n",
-       "                             \"learning_rate\": 1e-08, \"log_completions\": false,                                     \n",
-       "                             \"max_completion_length\": 128, \"max_grad_norm\": 0.05, \"max_steps\":                     \n",
-       "                             50, \"model_id\": \"HuggingFaceTB/SmolLM3-3B\",                                           \n",
-       "                             \"num_completions_to_print\": 8, \"num_generations\": 8,                                  \n",
-       "                             \"per_device_train_batch_size\": 2, \"remove_invalid_values\": false,                     \n",
-       "                             \"renormalize_logits\": true, \"repetition_penalty\": 1.05,                               \n",
-       "                             \"reward_weights\": \"1.0,0.5\", \"stop_on_collapse\": true,                                \n",
-       "                             \"temperature\": 0.7, \"top_k\": 0, \"top_p\": 1.0}, \"input_data_config\":                   \n",
-       "                             {\"code\": {\"TrainingInputMode\": \"File\", \"S3DistributionType\":                          \n",
-       "                             \"FullyReplicated\", \"RecordWrapperType\": \"None\"}, \"sm_drivers\":                        \n",
-       "                             {\"TrainingInputMode\": \"File\", \"S3DistributionType\":                                   \n",
-       "                             \"FullyReplicated\", \"RecordWrapperType\": \"None\"}, \"train\":                             \n",
-       "                             {\"TrainingInputMode\": \"File\", \"S3DistributionType\":                                   \n",
-       "                             \"FullyReplicated\", \"RecordWrapperType\": \"None\"}},                                     \n",
-       "                             \"input_config_dir\": \"/opt/ml/input/config\", \"input_data_dir\":                         \n",
-       "                             \"/opt/ml/input/data\", \"input_dir\": \"/opt/ml/input\", \"job_name\":                       \n",
-       "                             \"smolm3-grpo-toolcall-20260626-085316-20260626105316\", \"log_level\":                   \n",
-       "                             20, \"model_dir\": \"/opt/ml/model\", \"network_interface_name\": \"eth0\",                   \n",
-       "                             \"num_cpus\": 96, \"num_gpus\": 8, \"num_neurons\": 0, \"output_data_dir\":                   \n",
-       "                             \"/opt/ml/output/data\", \"resource_config\": {\"current_host\":                            \n",
-       "                             \"algo-1\", \"current_instance_type\": \"ml.p4d.24xlarge\",                                 \n",
-       "                             \"current_group_name\": \"homogeneousCluster\", \"hosts\": [\"algo-1\"],                      \n",
-       "                             \"instance_groups\": [{\"instance_group_name\": \"homogeneousCluster\",                     \n",
-       "                             \"instance_type\": \"ml.p4d.24xlarge\", \"hosts\": [\"algo-1\"]}],                            \n",
-       "                             \"network_interface_name\": \"eth0\", \"topology\": null}}                                  \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028843;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028844;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         SM_TRAINING_ENV=\u001b[1m{\u001b[0m\"channel_input_dirs\": \u001b[1m{\u001b[0m\"code\":                     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"/opt/ml/input/data/code\", \"sm_drivers\":                            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"/opt/ml/input/data/sm_drivers\", \"train\":                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"/opt/ml/input/data/train\"\u001b[1m}\u001b[0m, \"current_host\": \"algo-1\",              \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"current_instance_type\": \"ml.p4d.24xlarge\", \"hosts\": \u001b[1m[\u001b[0m\"algo-1\"\u001b[1m]\u001b[0m,    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"master_addr\": \"algo-1\", \"master_port\": \u001b[1m7777\u001b[0m, \"hyperparameters\":    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m{\u001b[0m\"beta\": \u001b[1m0.1\u001b[0m, \"deepspeed\": \"ds_zero3.json\",                         \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"gradient_accumulation_steps\": \u001b[1m8\u001b[0m, \"gradient_checkpointing\": true,   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"learning_rate\": \u001b[1m1e-08\u001b[0m, \"log_completions\": false,                   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"max_completion_length\": \u001b[1m128\u001b[0m, \"max_grad_norm\": \u001b[1m0.05\u001b[0m, \"max_steps\":   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m50\u001b[0m, \"model_id\": \"HuggingFaceTB/SmolLM3-3B\",                         \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"num_completions_to_print\": \u001b[1m8\u001b[0m, \"num_generations\": \u001b[1m8\u001b[0m,                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"per_device_train_batch_size\": \u001b[1m2\u001b[0m, \"remove_invalid_values\": false,   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"renormalize_logits\": true, \"repetition_penalty\": \u001b[1m1.05\u001b[0m,             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"reward_weights\": \"1.0,0.5\", \"stop_on_collapse\": true,              \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"temperature\": \u001b[1m0.7\u001b[0m, \"top_k\": \u001b[1m0\u001b[0m, \"top_p\": \u001b[1m1.0\u001b[0m\u001b[1m}\u001b[0m, \"input_data_config\": \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m{\u001b[0m\"code\": \u001b[1m{\u001b[0m\"TrainingInputMode\": \"File\", \"S3DistributionType\":        \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"FullyReplicated\", \"RecordWrapperType\": \"None\"\u001b[1m}\u001b[0m, \"sm_drivers\":      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m{\u001b[0m\"TrainingInputMode\": \"File\", \"S3DistributionType\":                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"FullyReplicated\", \"RecordWrapperType\": \"None\"\u001b[1m}\u001b[0m, \"train\":           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m{\u001b[0m\"TrainingInputMode\": \"File\", \"S3DistributionType\":                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"FullyReplicated\", \"RecordWrapperType\": \"None\"\u001b[1m}\u001b[0m\u001b[1m}\u001b[0m,                   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"input_config_dir\": \"/opt/ml/input/config\", \"input_data_dir\":       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"/opt/ml/input/data\", \"input_dir\": \"/opt/ml/input\", \"job_name\":     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"smolm3-grpo-toolcall-20260626-085316-20260626105316\", \"log_level\": \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m20\u001b[0m, \"model_dir\": \"/opt/ml/model\", \"network_interface_name\": \"eth0\", \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"num_cpus\": \u001b[1m96\u001b[0m, \"num_gpus\": \u001b[1m8\u001b[0m, \"num_neurons\": \u001b[1m0\u001b[0m, \"output_data_dir\": \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"/opt/ml/output/data\", \"resource_config\": \u001b[1m{\u001b[0m\"current_host\":          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"algo-1\", \"current_instance_type\": \"ml.p4d.24xlarge\",               \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"current_group_name\": \"homogeneousCluster\", \"hosts\": \u001b[1m[\u001b[0m\"algo-1\"\u001b[1m]\u001b[0m,    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"instance_groups\": \u001b[1m[\u001b[0m\u001b[1m{\u001b[0m\"instance_group_name\": \"homogeneousCluster\",   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"instance_type\": \"ml.p4d.24xlarge\", \"hosts\": \u001b[1m[\u001b[0m\"algo-1\"\u001b[1m]\u001b[0m\u001b[1m}\u001b[0m\u001b[1m]\u001b[0m,          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"network_interface_name\": \"eth0\", \"topology\": null\u001b[1m}\u001b[0m\u001b[1m}\u001b[0m                \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             ++ set +x                                                                             \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028851;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028852;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         ++ set +x                                                           \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             ++ cd /opt/ml/input/data/code                                                         \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028859;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028860;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         ++ cd /opt/ml/input/data/code                                       \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             Running Torchrun Driver                                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028867;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028868;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Running Torchrun Driver                                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             ++ echo 'Running Torchrun Driver'                                                     \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028875;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028876;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         ++ echo 'Running Torchrun Driver'                                   \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             ++ /opt/venv/bin/python3                                                              \n",
-       "                             /opt/ml/input/data/sm_drivers/distributed_drivers/torchrun_driver.p                   \n",
-       "                             y                                                                                     \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028883;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028884;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         ++ /opt/venv/bin/python3                                            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         /opt/ml/input/data/sm_drivers/distributed_drivers/torchrun_driver.p \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         y                                                                   \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             Executing command: torchrun --nnodes=1 --nproc_per_node=8 train.py                    \n",
-       "                             --beta 0.1 --deepspeed ds_zero3.json --gradient_accumulation_steps                    \n",
-       "                             8 --gradient_checkpointing true --learning_rate 1e-08                                 \n",
-       "                             --log_completions false --max_completion_length 128 --max_grad_norm                   \n",
-       "                             0.05 --max_steps 50 --model_id HuggingFaceTB/SmolLM3-3B                               \n",
-       "                             --num_completions_to_print 8 --num_generations 8                                      \n",
-       "                             --per_device_train_batch_size 2 --remove_invalid_values false                         \n",
-       "                             --renormalize_logits true --repetition_penalty 1.05                                   \n",
-       "                             --reward_weights 1.0,0.5 --stop_on_collapse true --temperature 0.7                    \n",
-       "                             --top_k 0 --top_p 1.0                                                                 \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028891;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028892;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Executing command: torchrun --nnodes=\u001b[1m1\u001b[0m --nproc_per_node=\u001b[1m8\u001b[0m train.py  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         --beta \u001b[1m0.1\u001b[0m --deepspeed ds_zero3.json --gradient_accumulation_steps  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m8\u001b[0m --gradient_checkpointing true --learning_rate \u001b[1m1e-08\u001b[0m               \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         --log_completions false --max_completion_length \u001b[1m128\u001b[0m --max_grad_norm \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m0.05\u001b[0m --max_steps \u001b[1m50\u001b[0m --model_id HuggingFaceTB/SmolLM3-3B             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         --num_completions_to_print \u001b[1m8\u001b[0m --num_generations \u001b[1m8\u001b[0m                    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         --per_device_train_batch_size \u001b[1m2\u001b[0m --remove_invalid_values false       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         --renormalize_logits true --repetition_penalty \u001b[1m1.05\u001b[0m                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         --reward_weights \u001b[1m1.0\u001b[0m,\u001b[1m0.5\u001b[0m --stop_on_collapse true --temperature \u001b[1m0.7\u001b[0m  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         --top_k \u001b[1m0\u001b[0m --top_p \u001b[1m1.0\u001b[0m                                               \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             W0626 09:01:36.952000 89 torch/distributed/run.py:851]                                \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028899;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028900;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         W0626 09:01:36.\u001b[1m952000\u001b[0m \u001b[1m89\u001b[0m torch/distributed/run.py:\u001b[1m851\u001b[0m\u001b[1m]\u001b[0m              \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             W0626 09:01:36.952000 89 torch/distributed/run.py:851]                                \n",
-       "                             *****************************************                                             \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028907;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028908;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         W0626 09:01:36.\u001b[1m952000\u001b[0m \u001b[1m89\u001b[0m torch/distributed/run.py:\u001b[1m851\u001b[0m\u001b[1m]\u001b[0m              \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         *****************************************                           \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             W0626 09:01:36.952000 89 torch/distributed/run.py:851] Setting                        \n",
-       "                             OMP_NUM_THREADS environment variable for each process to be 1 in                      \n",
-       "                             default, to avoid your system being overloaded, please further tune                   \n",
-       "                             the variable for optimal performance in your application as needed.                   \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028915;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028916;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         W0626 09:01:36.\u001b[1m952000\u001b[0m \u001b[1m89\u001b[0m torch/distributed/run.py:\u001b[1m851\u001b[0m\u001b[1m]\u001b[0m Setting      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         OMP_NUM_THREADS environment variable for each process to be \u001b[1m1\u001b[0m in    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         default, to avoid your system being overloaded, please further tune \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         the variable for optimal performance in your application as needed. \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             W0626 09:01:36.952000 89 torch/distributed/run.py:851]                                \n",
-       "                             *****************************************                                             \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028923;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028924;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         W0626 09:01:36.\u001b[1m952000\u001b[0m \u001b[1m89\u001b[0m torch/distributed/run.py:\u001b[1m851\u001b[0m\u001b[1m]\u001b[0m              \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         *****************************************                           \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:02:16] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             df: /root/.triton/autotune: No such file or directory                                 \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:02:16]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028931;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028932;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         df: /root/.triton/autotune: No such file or directory               \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             NCCL version 2.28.9+cuda13.0                                                          \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028939;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028940;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         NCCL version \u001b[1m2.28\u001b[0m.\u001b[1m9\u001b[0m+cuda13.\u001b[1m0\u001b[0m                                        \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:02:27] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             Fetching 2 files:   0%|          | 0/2 [00:00<?, ?it/s]#015Fetching                   \n",
-       "                             2 files:   0%|          | 0/2 [00:00<?, ?it/s]#015Fetching 2 files:                   \n",
-       "                             0%|          | 0/2 [00:00<?, ?it/s]#015Fetching 2 files:   0%|                        \n",
-       "                             | 0/2 [00:00<?, ?it/s]#015Fetching 2 files:   0%|          | 0/2                      \n",
-       "                             [00:00<?, ?it/s]#015Fetching 2 files:   0%|          | 0/2                            \n",
-       "                             [00:00<?, ?it/s]#015Fetching 2 files:   0%|          | 0/2                            \n",
-       "                             [00:00<?, ?it/s]#015Fetching 2 files:   0%|          | 0/2                            \n",
-       "                             [00:00<?, ?it/s]#015Fetching 2 files:  50%|█████     | 1/2                            \n",
-       "                             [00:07<00:07,  7.44s/it]#015Fetching 2 files: 100%|██████████| 2/2                    \n",
-       "                             [00:07<00:00,  3.72s/it]                                                              \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:02:27]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028947;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028948;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:   \u001b[1m0\u001b[0m%|          | \u001b[1m0\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:00                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             Fetching 2 files:  50%|█████     | 1/2 [00:07<00:07,                                  \n",
-       "                             7.45s/it]#015Fetching 2 files: 100%|██████████| 2/2 [00:07<00:00,                     \n",
-       "                             3.72s/it]                                                                             \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028955;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028956;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:  \u001b[1m50\u001b[0m%|█████     | \u001b[1m1\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:07<00:07,                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m7.\u001b[0m45s/it\u001b[1m]\u001b[0m#015Fetching \u001b[1m2\u001b[0m files: \u001b[1m100\u001b[0m%|██████████| \u001b[1m2\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:07<00:00,   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m3.\u001b[0m72s/it\u001b[1m]\u001b[0m                                                           \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             Fetching 2 files:  50%|█████     | 1/2 [00:07<00:07,                                  \n",
-       "                             7.45s/it]#015Fetching 2 files: 100%|██████████| 2/2 [00:07<00:00,                     \n",
-       "                             3.72s/it]                                                                             \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028963;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028964;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:  \u001b[1m50\u001b[0m%|█████     | \u001b[1m1\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:07<00:07,                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m7.\u001b[0m45s/it\u001b[1m]\u001b[0m#015Fetching \u001b[1m2\u001b[0m files: \u001b[1m100\u001b[0m%|██████████| \u001b[1m2\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:07<00:00,   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m3.\u001b[0m72s/it\u001b[1m]\u001b[0m                                                           \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             Fetching 2 files:  50%|█████     | 1/2 [00:07<00:07,                                  \n",
-       "                             7.45s/it]#015Fetching 2 files: 100%|██████████| 2/2 [00:07<00:00,                     \n",
-       "                             3.73s/it]                                                                             \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028971;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028972;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:  \u001b[1m50\u001b[0m%|█████     | \u001b[1m1\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:07<00:07,                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m7.\u001b[0m45s/it\u001b[1m]\u001b[0m#015Fetching \u001b[1m2\u001b[0m files: \u001b[1m100\u001b[0m%|██████████| \u001b[1m2\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:07<00:00,   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m3.\u001b[0m73s/it\u001b[1m]\u001b[0m                                                           \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             Fetching 2 files:  50%|█████     | 1/2 [00:07<00:07,                                  \n",
-       "                             7.45s/it]#015Fetching 2 files: 100%|██████████| 2/2 [00:07<00:00,                     \n",
-       "                             3.72s/it]                                                                             \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028979;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028980;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:  \u001b[1m50\u001b[0m%|█████     | \u001b[1m1\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:07<00:07,                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m7.\u001b[0m45s/it\u001b[1m]\u001b[0m#015Fetching \u001b[1m2\u001b[0m files: \u001b[1m100\u001b[0m%|██████████| \u001b[1m2\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:07<00:00,   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m3.\u001b[0m72s/it\u001b[1m]\u001b[0m                                                           \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             Fetching 2 files:  50%|█████     | 1/2 [00:07<00:07,                                  \n",
-       "                             7.45s/it]#015Fetching 2 files: 100%|██████████| 2/2 [00:07<00:00,                     \n",
-       "                             3.72s/it]                                                                             \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028987;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028988;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:  \u001b[1m50\u001b[0m%|█████     | \u001b[1m1\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:07<00:07,                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m7.\u001b[0m45s/it\u001b[1m]\u001b[0m#015Fetching \u001b[1m2\u001b[0m files: \u001b[1m100\u001b[0m%|██████████| \u001b[1m2\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:07<00:00,   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m3.\u001b[0m72s/it\u001b[1m]\u001b[0m                                                           \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             Fetching 2 files:  50%|█████     | 1/2 [00:07<00:07,                                  \n",
-       "                             7.46s/it]#015Fetching 2 files: 100%|██████████| 2/2 [00:07<00:00,                     \n",
-       "                             3.73s/it]                                                                             \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17028995;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17028996;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:  \u001b[1m50\u001b[0m%|█████     | \u001b[1m1\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:07<00:07,                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m7.\u001b[0m46s/it\u001b[1m]\u001b[0m#015Fetching \u001b[1m2\u001b[0m files: \u001b[1m100\u001b[0m%|██████████| \u001b[1m2\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:07<00:00,   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m3.\u001b[0m73s/it\u001b[1m]\u001b[0m                                                           \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             Fetching 2 files:  50%|█████     | 1/2 [00:07<00:07,                                  \n",
-       "                             7.46s/it]#015Fetching 2 files: 100%|██████████| 2/2 [00:07<00:00,                     \n",
-       "                             3.73s/it]                                                                             \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029003;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029004;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:  \u001b[1m50\u001b[0m%|█████     | \u001b[1m1\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:07<00:07,                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m7.\u001b[0m46s/it\u001b[1m]\u001b[0m#015Fetching \u001b[1m2\u001b[0m files: \u001b[1m100\u001b[0m%|██████████| \u001b[1m2\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:07<00:00,   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m3.\u001b[0m73s/it\u001b[1m]\u001b[0m                                                           \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             Fetching 2 files:   0%|          | 0/2 [00:00<?, ?it/s]#015Fetching                   \n",
-       "                             2 files: 100%|██████████| 2/2 [00:00<00:00, 39199.10it/s]                             \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029011;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029012;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:   \u001b[1m0\u001b[0m%|          | \u001b[1m0\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:00                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             Fetching 2 files:   0%|          | 0/2 [00:00<?, ?it/s]#015Fetching                   \n",
-       "                             2 files: 100%|██████████| 2/2 [00:00<00:00, 36314.32it/s]                             \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029019;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029020;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:   \u001b[1m0\u001b[0m%|          | \u001b[1m0\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:00                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             Fetching 2 files:   0%|          | 0/2 [00:00<?, ?it/s]#015Fetching                   \n",
-       "                             2 files: 100%|██████████| 2/2 [00:00<00:00, 41120.63it/s]                             \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029027;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029028;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:   \u001b[1m0\u001b[0m%|          | \u001b[1m0\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:00                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             Fetching 2 files:   0%|          | 0/2 [00:00<?, ?it/s]#015Fetching                   \n",
-       "                             2 files: 100%|██████████| 2/2 [00:00<00:00, 41734.37it/s]                             \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029035;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029036;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:   \u001b[1m0\u001b[0m%|          | \u001b[1m0\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:00                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             Fetching 2 files:   0%|          | 0/2 [00:00<?, ?it/s]#015Fetching                   \n",
-       "                             2 files: 100%|██████████| 2/2 [00:00<00:00, 41323.19it/s]                             \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029043;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029044;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:   \u001b[1m0\u001b[0m%|          | \u001b[1m0\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:00                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             Fetching 2 files:   0%|          | 0/2 [00:00<?, ?it/s]#015Fetching                   \n",
-       "                             2 files: 100%|██████████| 2/2 [00:00<00:00, 25497.29it/s]                             \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029051;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029052;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:   \u001b[1m0\u001b[0m%|          | \u001b[1m0\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:00                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             Fetching 2 files:   0%|          | 0/2 [00:00<?, ?it/s]#015Fetching                   \n",
-       "                             2 files: 100%|██████████| 2/2 [00:00<00:00, 39756.44it/s]                             \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029059;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029060;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:   \u001b[1m0\u001b[0m%|          | \u001b[1m0\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:00                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             Fetching 2 files:   0%|          | 0/2 [00:00<?, ?it/s]#015Fetching                   \n",
-       "                             2 files: 100%|██████████| 2/2 [00:00<00:00, 40721.40it/s]                             \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029067;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029068;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Fetching \u001b[1m2\u001b[0m files:   \u001b[1m0\u001b[0m%|          | \u001b[1m0\u001b[0m/\u001b[1m2\u001b[0m \u001b[1m[\u001b[0m00:00[06/26/26 11:02:33] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                              The tokenizer has new PAD/BOS/EOS tokens that differ from the                        \n",
-       "                             model config and generation config. The model config and generation                   \n",
-       "                             config were aligned accordingly, being updated with the tokenizer's                   \n",
-       "                             values. Updated tokens: {'bos_token_id': None, 'pad_token_id':                        \n",
-       "                             128012}.                                                                              \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:02:33]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029075;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029076;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m          The tokenizer has new PAD/BOS/EOS tokens that differ from the      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         model config and generation config. The model config and generation \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         config were aligned accordingly, being updated with the tokenizer's \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         values. Updated tokens: \u001b[1m{\u001b[0m'bos_token_id': \u001b[3mNone\u001b[0m, 'pad_token_id':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m128012\u001b[0m\u001b[1m}\u001b[0m.                                                            \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                              The tokenizer has new PAD/BOS/EOS tokens that differ from the                        \n",
-       "                             model config and generation config. The model config and generation                   \n",
-       "                             config were aligned accordingly, being updated with the tokenizer's                   \n",
-       "                             values. Updated tokens: {'bos_token_id': None, 'pad_token_id':                        \n",
-       "                             128012}.                                                                              \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029083;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029084;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m          The tokenizer has new PAD/BOS/EOS tokens that differ from the      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         model config and generation config. The model config and generation \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         config were aligned accordingly, being updated with the tokenizer's \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         values. Updated tokens: \u001b[1m{\u001b[0m'bos_token_id': \u001b[3mNone\u001b[0m, 'pad_token_id':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m128012\u001b[0m\u001b[1m}\u001b[0m.                                                            \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                              The tokenizer has new PAD/BOS/EOS tokens that differ from the                        \n",
-       "                             model config and generation config. The model config and generation                   \n",
-       "                             config were aligned accordingly, being updated with the tokenizer's                   \n",
-       "                             values. Updated tokens: {'bos_token_id': None, 'pad_token_id':                        \n",
-       "                             128012}.                                                                              \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029091;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029092;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m          The tokenizer has new PAD/BOS/EOS tokens that differ from the      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         model config and generation config. The model config and generation \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         config were aligned accordingly, being updated with the tokenizer's \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         values. Updated tokens: \u001b[1m{\u001b[0m'bos_token_id': \u001b[3mNone\u001b[0m, 'pad_token_id':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m128012\u001b[0m\u001b[1m}\u001b[0m.                                                            \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                              The tokenizer has new PAD/BOS/EOS tokens that differ from the                        \n",
-       "                             model config and generation config. The model config and generation                   \n",
-       "                             config were aligned accordingly, being updated with the tokenizer's                   \n",
-       "                             values. Updated tokens: {'bos_token_id': None, 'pad_token_id':                        \n",
-       "                             128012}.                                                                              \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029099;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029100;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m          The tokenizer has new PAD/BOS/EOS tokens that differ from the      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         model config and generation config. The model config and generation \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         config were aligned accordingly, being updated with the tokenizer's \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         values. Updated tokens: \u001b[1m{\u001b[0m'bos_token_id': \u001b[3mNone\u001b[0m, 'pad_token_id':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m128012\u001b[0m\u001b[1m}\u001b[0m.                                                            \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                              The tokenizer has new PAD/BOS/EOS tokens that differ from the                        \n",
-       "                             model config and generation config. The model config and generation                   \n",
-       "                             config were aligned accordingly, being updated with the tokenizer's                   \n",
-       "                             values. Updated tokens: {'bos_token_id': None, 'pad_token_id':                        \n",
-       "                             128012}.                                                                              \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029107;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029108;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m          The tokenizer has new PAD/BOS/EOS tokens that differ from the      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         model config and generation config. The model config and generation \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         config were aligned accordingly, being updated with the tokenizer's \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         values. Updated tokens: \u001b[1m{\u001b[0m'bos_token_id': \u001b[3mNone\u001b[0m, 'pad_token_id':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m128012\u001b[0m\u001b[1m}\u001b[0m.                                                            \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                              The tokenizer has new PAD/BOS/EOS tokens that differ from the                        \n",
-       "                             model config and generation config. The model config and generation                   \n",
-       "                             config were aligned accordingly, being updated with the tokenizer's                   \n",
-       "                             values. Updated tokens: {'bos_token_id': None, 'pad_token_id':                        \n",
-       "                             128012}.                                                                              \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029115;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029116;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m          The tokenizer has new PAD/BOS/EOS tokens that differ from the      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         model config and generation config. The model config and generation \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         config were aligned accordingly, being updated with the tokenizer's \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         values. Updated tokens: \u001b[1m{\u001b[0m'bos_token_id': \u001b[3mNone\u001b[0m, 'pad_token_id':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m128012\u001b[0m\u001b[1m}\u001b[0m.                                                            \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                              The tokenizer has new PAD/BOS/EOS tokens that differ from the                        \n",
-       "                             model config and generation config. The model config and generation                   \n",
-       "                             config were aligned accordingly, being updated with the tokenizer's                   \n",
-       "                             values. Updated tokens: {'bos_token_id': None, 'pad_token_id':                        \n",
-       "                             128012}.                                                                              \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029123;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029124;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m          The tokenizer has new PAD/BOS/EOS tokens that differ from the      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         model config and generation config. The model config and generation \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         config were aligned accordingly, being updated with the tokenizer's \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         values. Updated tokens: \u001b[1m{\u001b[0m'bos_token_id': \u001b[3mNone\u001b[0m, 'pad_token_id':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m128012\u001b[0m\u001b[1m}\u001b[0m.                                                            \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                              The tokenizer has new PAD/BOS/EOS tokens that differ from the                        \n",
-       "                             model config and generation config. The model config and generation                   \n",
-       "                             config were aligned accordingly, being updated with the tokenizer's                   \n",
-       "                             values. Updated tokens: {'bos_token_id': None, 'pad_token_id':                        \n",
-       "                             128012}.                                                                              \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029131;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029132;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m          The tokenizer has new PAD/BOS/EOS tokens that differ from the      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         model config and generation config. The model config and generation \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         config were aligned accordingly, being updated with the tokenizer's \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         values. Updated tokens: \u001b[1m{\u001b[0m'bos_token_id': \u001b[3mNone\u001b[0m, 'pad_token_id':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m128012\u001b[0m\u001b[1m}\u001b[0m.                                                            \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             [RANK 0] Gradient accumulation steps mismatch:                                        \n",
-       "                             GradientAccumulationPlugin has 1, DeepSpeed config has 8. Using                       \n",
-       "                             DeepSpeed's value.                                                                    \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029139;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029140;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m[\u001b[0mRANK \u001b[1m0\u001b[0m\u001b[1m]\u001b[0m Gradient accumulation steps mismatch:                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         GradientAccumulationPlugin has \u001b[1m1\u001b[0m, DeepSpeed config has \u001b[1m8\u001b[0m. Using     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         DeepSpeed's value.                                                  \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:02:43] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             0%|          | 0/50 [00:00<?, ?it/s] Ignoring                                         \n",
-       "                             clean_up_tokenization_spaces=True for BPE tokenizer                                   \n",
-       "                             TokenizersBackend. The clean_up_tokenization post-processing step                     \n",
-       "                             is designed for WordPiece tokenizers and is destructive for BPE (it                   \n",
-       "                             strips spaces before punctuation). Set                                                \n",
-       "                             clean_up_tokenization_spaces=False to suppress this warning, or set                   \n",
-       "                             clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou                   \n",
-       "                             tput=True to force cleanup anyway.                                                    \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:02:43]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029147;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029148;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m0\u001b[0m%|          | \u001b[1m0\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m00:00                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                              Ignoring clean_up_tokenization_spaces=True for BPE tokenizer                         \n",
-       "                             TokenizersBackend. The clean_up_tokenization post-processing step                     \n",
-       "                             is designed for WordPiece tokenizers and is destructive for BPE (it                   \n",
-       "                             strips spaces before punctuation). Set                                                \n",
-       "                             clean_up_tokenization_spaces=False to suppress this warning, or set                   \n",
-       "                             clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou                   \n",
-       "                             tput=True to force cleanup anyway.                                                    \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029155;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029156;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m          Ignoring clean_up_tokenization_spaces=\u001b[3mTrue\u001b[0m for BPE tokenizer       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         TokenizersBackend. The clean_up_tokenization post-processing step   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         is designed for WordPiece tokenizers and is destructive for BPE \u001b[1m(\u001b[0mit \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         strips spaces before punctuation\u001b[1m)\u001b[0m. Set                              \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         clean_up_tokenization_spaces=\u001b[3mFalse\u001b[0m to suppress this warning, or set \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         tput=\u001b[3mTrue\u001b[0m to force cleanup anyway.                                  \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                              Ignoring clean_up_tokenization_spaces=True for BPE tokenizer                         \n",
-       "                             TokenizersBackend. The clean_up_tokenization post-processing step                     \n",
-       "                             is designed for WordPiece tokenizers and is destructive for BPE (it                   \n",
-       "                             strips spaces before punctuation). Set                                                \n",
-       "                             clean_up_tokenization_spaces=False to suppress this warning, or set                   \n",
-       "                             clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou                   \n",
-       "                             tput=True to force cleanup anyway.                                                    \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029163;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029164;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m          Ignoring clean_up_tokenization_spaces=\u001b[3mTrue\u001b[0m for BPE tokenizer       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         TokenizersBackend. The clean_up_tokenization post-processing step   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         is designed for WordPiece tokenizers and is destructive for BPE \u001b[1m(\u001b[0mit \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         strips spaces before punctuation\u001b[1m)\u001b[0m. Set                              \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         clean_up_tokenization_spaces=\u001b[3mFalse\u001b[0m to suppress this warning, or set \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         tput=\u001b[3mTrue\u001b[0m to force cleanup anyway.                                  \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                              Ignoring clean_up_tokenization_spaces=True for BPE tokenizer                         \n",
-       "                             TokenizersBackend. The clean_up_tokenization post-processing step                     \n",
-       "                             is designed for WordPiece tokenizers and is destructive for BPE (it                   \n",
-       "                             strips spaces before punctuation). Set                                                \n",
-       "                             clean_up_tokenization_spaces=False to suppress this warning, or set                   \n",
-       "                             clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou                   \n",
-       "                             tput=True to force cleanup anyway.                                                    \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029171;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029172;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m          Ignoring clean_up_tokenization_spaces=\u001b[3mTrue\u001b[0m for BPE tokenizer       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         TokenizersBackend. The clean_up_tokenization post-processing step   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         is designed for WordPiece tokenizers and is destructive for BPE \u001b[1m(\u001b[0mit \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         strips spaces before punctuation\u001b[1m)\u001b[0m. Set                              \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         clean_up_tokenization_spaces=\u001b[3mFalse\u001b[0m to suppress this warning, or set \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         tput=\u001b[3mTrue\u001b[0m to force cleanup anyway.                                  \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                              Ignoring clean_up_tokenization_spaces=True for BPE tokenizer                         \n",
-       "                             TokenizersBackend. The clean_up_tokenization post-processing step                     \n",
-       "                             is designed for WordPiece tokenizers and is destructive for BPE (it                   \n",
-       "                             strips spaces before punctuation). Set                                                \n",
-       "                             clean_up_tokenization_spaces=False to suppress this warning, or set                   \n",
-       "                             clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou                   \n",
-       "                             tput=True to force cleanup anyway.                                                    \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029179;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029180;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m          Ignoring clean_up_tokenization_spaces=\u001b[3mTrue\u001b[0m for BPE tokenizer       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         TokenizersBackend. The clean_up_tokenization post-processing step   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         is designed for WordPiece tokenizers and is destructive for BPE \u001b[1m(\u001b[0mit \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         strips spaces before punctuation\u001b[1m)\u001b[0m. Set                              \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         clean_up_tokenization_spaces=\u001b[3mFalse\u001b[0m to suppress this warning, or set \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         tput=\u001b[3mTrue\u001b[0m to force cleanup anyway.                                  \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                              Ignoring clean_up_tokenization_spaces=True for BPE tokenizer                         \n",
-       "                             TokenizersBackend. The clean_up_tokenization post-processing step                     \n",
-       "                             is designed for WordPiece tokenizers and is destructive for BPE (it                   \n",
-       "                             strips spaces before punctuation). Set                                                \n",
-       "                             clean_up_tokenization_spaces=False to suppress this warning, or set                   \n",
-       "                             clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou                   \n",
-       "                             tput=True to force cleanup anyway.                                                    \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029187;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029188;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m          Ignoring clean_up_tokenization_spaces=\u001b[3mTrue\u001b[0m for BPE tokenizer       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         TokenizersBackend. The clean_up_tokenization post-processing step   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         is designed for WordPiece tokenizers and is destructive for BPE \u001b[1m(\u001b[0mit \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         strips spaces before punctuation\u001b[1m)\u001b[0m. Set                              \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         clean_up_tokenization_spaces=\u001b[3mFalse\u001b[0m to suppress this warning, or set \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         tput=\u001b[3mTrue\u001b[0m to force cleanup anyway.                                  \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                              Ignoring clean_up_tokenization_spaces=True for BPE tokenizer                         \n",
-       "                             TokenizersBackend. The clean_up_tokenization post-processing step                     \n",
-       "                             is designed for WordPiece tokenizers and is destructive for BPE (it                   \n",
-       "                             strips spaces before punctuation). Set                                                \n",
-       "                             clean_up_tokenization_spaces=False to suppress this warning, or set                   \n",
-       "                             clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou                   \n",
-       "                             tput=True to force cleanup anyway.                                                    \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029195;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029196;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m          Ignoring clean_up_tokenization_spaces=\u001b[3mTrue\u001b[0m for BPE tokenizer       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         TokenizersBackend. The clean_up_tokenization post-processing step   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         is designed for WordPiece tokenizers and is destructive for BPE \u001b[1m(\u001b[0mit \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         strips spaces before punctuation\u001b[1m)\u001b[0m. Set                              \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         clean_up_tokenization_spaces=\u001b[3mFalse\u001b[0m to suppress this warning, or set \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         tput=\u001b[3mTrue\u001b[0m to force cleanup anyway.                                  \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                              Ignoring clean_up_tokenization_spaces=True for BPE tokenizer                         \n",
-       "                             TokenizersBackend. The clean_up_tokenization post-processing step                     \n",
-       "                             is designed for WordPiece tokenizers and is destructive for BPE (it                   \n",
-       "                             strips spaces before punctuation). Set                                                \n",
-       "                             clean_up_tokenization_spaces=False to suppress this warning, or set                   \n",
-       "                             clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou                   \n",
-       "                             tput=True to force cleanup anyway.                                                    \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029203;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029204;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m          Ignoring clean_up_tokenization_spaces=\u001b[3mTrue\u001b[0m for BPE tokenizer       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         TokenizersBackend. The clean_up_tokenization post-processing step   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         is designed for WordPiece tokenizers and is destructive for BPE \u001b[1m(\u001b[0mit \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         strips spaces before punctuation\u001b[1m)\u001b[0m. Set                              \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         clean_up_tokenization_spaces=\u001b[3mFalse\u001b[0m to suppress this warning, or set \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         clean_up_tokenization_spaces_for_bpe_even_though_it_will_corrupt_ou \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         tput=\u001b[3mTrue\u001b[0m to force cleanup anyway.                                  \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:02:54] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             2%|▏         | 1/50 [00:20<16:25, 20.10s/it]#015                                      \n",
-       "                             #015{'loss': '0.03134', 'grad_norm': '2.996', 'learning_rate':                        \n",
-       "                             '1e-08', 'num_tokens': '6.144e+04', 'completions/mean_length':                        \n",
-       "                             '30.66', 'completions/min_length': '17', 'completions/max_length':                    \n",
-       "                             '128', 'completions/clipped_ratio': '0.01562',                                        \n",
-       "                             'completions/mean_terminated_length': '29.12',                                        \n",
-       "                             'completions/min_terminated_length': '17',                                            \n",
-       "                             'completions/max_terminated_length': '104',                                           \n",
-       "                             'rewards/tool_call_reward/mean': '0.8281',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.3788',                                             \n",
-       "                             'rewards/format_reward/mean': '0.937', 'rewards/format_reward/std':                   \n",
-       "                             '0.2449', 'reward': '1.297', 'reward_std': '0.4594',                                  \n",
-       "                             'frac_reward_zero_std': '0', 'kl': '4.555e-06', 'entropy':                            \n",
-       "                             '0.02574', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',                     \n",
-       "                             'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',                              \n",
-       "                             'clip_ratio/region_mean': '0', 'step_time': '20.09', 'epoch':                         \n",
-       "                             '0.008'}                                                                              \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:02:54]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029211;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029212;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m2\u001b[0m%|▏         | \u001b[1m1\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m00:20<16:25, \u001b[1m20.\u001b[0m10s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.03134', 'grad_norm': '2.996', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '1e-08', 'num_tokens': '6.144e+04', 'completions/mean_length':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '30.66', 'completions/min_length': '17', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '128', 'completions/clipped_ratio': '0.01562',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '29.12',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '17',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '104',                         \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8281',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.3788',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.937', 'rewards/format_reward/std': \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.2449', 'reward': '1.297', 'reward_std': '0.4594',                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'frac_reward_zero_std': '0', 'kl': '4.555e-06', 'entropy':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.02574', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/region_mean': '0', 'step_time': '20.09', 'epoch':       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.008'\u001b[1m}\u001b[0m                                                            \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:03:10] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             2%|▏         | 1/50 [00:20<16:25, 20.10s/it]#015  4%|▍         |                      \n",
-       "                             2/50 [00:34<13:37, 17.04s/it]#015                                                     \n",
-       "                             #015{'loss': '0.002893', 'grad_norm': '2.016', 'learning_rate':                       \n",
-       "                             '9.8e-09', 'num_tokens': '1.326e+05', 'completions/mean_length':                      \n",
-       "                             '36.78', 'completions/min_length': '20', 'completions/max_length':                    \n",
-       "                             '82', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '36.78',                                        \n",
-       "                             'completions/min_terminated_length': '20',                                            \n",
-       "                             'completions/max_terminated_length': '82',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.6484',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4793',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9371',                                               \n",
-       "                             'rewards/format_reward/std': '0.2445', 'reward': '1.117',                             \n",
-       "                             'reward_std': '0.5346', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '8.694e-05', 'entropy': '0.01816', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '14.88', 'epoch': '0.016'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:03:10]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029219;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029220;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m2\u001b[0m%|▏         | \u001b[1m1\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m00:20<16:25, \u001b[1m20.\u001b[0m10s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m  \u001b[1m4\u001b[0m%|▍         |    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m2\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m00:34<13:37, \u001b[1m17.\u001b[0m04s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.002893', 'grad_norm': '2.016', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '9.8e-09', 'num_tokens': '1.326e+05', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '36.78', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '82', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '36.78',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '82',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.6484',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4793',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9371',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2445', 'reward': '1.117',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5346', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '8.694e-05', 'entropy': '0.01816', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '14.88', 'epoch': '0.016'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:03:26] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             4%|▍         | 2/50 [00:34<13:37, 17.04s/it]#015  6%|▌         |                      \n",
-       "                             3/50 [00:51<12:59, 16.58s/it]#015                                                     \n",
-       "                             #015{'loss': '-0.0008295', 'grad_norm': '2.204', 'learning_rate':                     \n",
-       "                             '9.6e-09', 'num_tokens': '2.039e+05', 'completions/mean_length':                      \n",
-       "                             '40.09', 'completions/min_length': '20', 'completions/max_length':                    \n",
-       "                             '84', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '40.09',                                        \n",
-       "                             'completions/min_terminated_length': '20',                                            \n",
-       "                             'completions/max_terminated_length': '84',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7656',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4253',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9292',                                               \n",
-       "                             'rewards/format_reward/std': '0.2583', 'reward': '1.23',                              \n",
-       "                             'reward_std': '0.5021', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.0002372', 'entropy': '0.02427', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '16.03', 'epoch': '0.024'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:03:26]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029227;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029228;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m4\u001b[0m%|▍         | \u001b[1m2\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m00:34<13:37, \u001b[1m17.\u001b[0m04s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m  \u001b[1m6\u001b[0m%|▌         |    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m3\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m00:51<12:59, \u001b[1m16.\u001b[0m58s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.0008295', 'grad_norm': '2.204', 'learning_rate':   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '9.6e-09', 'num_tokens': '2.039e+05', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '40.09', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '84', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '40.09',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '84',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7656',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4253',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9292',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2583', 'reward': '1.23',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5021', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0002372', 'entropy': '0.02427', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '16.03', 'epoch': '0.024'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:03:42] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             6%|▌         | 3/50 [00:51<12:59, 16.58s/it]#015  8%|▊         |                      \n",
-       "                             4/50 [01:04<11:53, 15.52s/it]#015                                                     \n",
-       "                             #015{'loss': '0.007161', 'grad_norm': '1.699', 'learning_rate':                       \n",
-       "                             '9.4e-09', 'num_tokens': '2.644e+05', 'completions/mean_length':                      \n",
-       "                             '34.21', 'completions/min_length': '22', 'completions/max_length':                    \n",
-       "                             '57', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '34.21',                                        \n",
-       "                             'completions/min_terminated_length': '22',                                            \n",
-       "                             'completions/max_terminated_length': '57',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.8359',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.3718',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9214',                                               \n",
-       "                             'rewards/format_reward/std': '0.2711', 'reward': '1.297',                             \n",
-       "                             'reward_std': '0.472', 'frac_reward_zero_std': '0', 'kl':                             \n",
-       "                             '6.244e-06', 'entropy': '0.01326', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '13.87', 'epoch': '0.032'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:03:42]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029235;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029236;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m6\u001b[0m%|▌         | \u001b[1m3\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m00:51<12:59, \u001b[1m16.\u001b[0m58s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m  \u001b[1m8\u001b[0m%|▊         |    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m4\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:04<11:53, \u001b[1m15.\u001b[0m52s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.007161', 'grad_norm': '1.699', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '9.4e-09', 'num_tokens': '2.644e+05', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '34.21', 'completions/min_length': '22', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '57', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '34.21',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '22',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '57',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8359',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.3718',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9214',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2711', 'reward': '1.297',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.472', 'frac_reward_zero_std': '0', 'kl':           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '6.244e-06', 'entropy': '0.01326', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '13.87', 'epoch': '0.032'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:03:53] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             8%|▊         | 4/50 [01:04<11:53, 15.52s/it]#015 10%|█         |                      \n",
-       "                             5/50 [01:17<10:42, 14.28s/it]#015                                                     \n",
-       "                             #015{'loss': '-0.004775', 'grad_norm': '2.619', 'learning_rate':                      \n",
-       "                             '9.2e-09', 'num_tokens': '3.378e+05', 'completions/mean_length':                      \n",
-       "                             '35.8', 'completions/min_length': '24', 'completions/max_length':                     \n",
-       "                             '55', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '35.8',                                         \n",
-       "                             'completions/min_terminated_length': '24',                                            \n",
-       "                             'completions/max_terminated_length': '55',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7734',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4203',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9371',                                               \n",
-       "                             'rewards/format_reward/std': '0.2444', 'reward': '1.242',                             \n",
-       "                             'reward_std': '0.4905', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '7.967e-05', 'entropy': '0.01727', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '12.06', 'epoch': '0.04'}                                                \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:03:53]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029243;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029244;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m8\u001b[0m%|▊         | \u001b[1m4\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:04<11:53, \u001b[1m15.\u001b[0m52s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m10\u001b[0m%|█         |    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m5\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:17<10:42, \u001b[1m14.\u001b[0m28s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.004775', 'grad_norm': '2.619', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '9.2e-09', 'num_tokens': '3.378e+05', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '35.8', 'completions/min_length': '24', 'completions/max_length':   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '55', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '35.8',                       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '24',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '55',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7734',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4203',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9371',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2444', 'reward': '1.242',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.4905', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '7.967e-05', 'entropy': '0.01727', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '12.06', 'epoch': '0.04'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:04:04] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             10%|█         | 5/50 [01:17<10:42, 14.28s/it]#015 12%|█▏        |                     \n",
-       "                             6/50 [01:29<09:56, 13.55s/it]#015                                                     \n",
-       "                             #015{'loss': '8.66e-07', 'grad_norm': '1.851', 'learning_rate':                       \n",
-       "                             '9e-09', 'num_tokens': '4.018e+05', 'completions/mean_length':                        \n",
-       "                             '28.62', 'completions/min_length': '20', 'completions/max_length':                    \n",
-       "                             '44', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '28.62',                                        \n",
-       "                             'completions/min_terminated_length': '20',                                            \n",
-       "                             'completions/max_terminated_length': '44',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.8516',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.3569',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9136',                                               \n",
-       "                             'rewards/format_reward/std': '0.2828', 'reward': '1.308',                             \n",
-       "                             'reward_std': '0.4707', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '6.785e-06', 'entropy': '0.01261', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '12.11', 'epoch': '0.048'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:04:04]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029251;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029252;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m10\u001b[0m%|█         | \u001b[1m5\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:17<10:42, \u001b[1m14.\u001b[0m28s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m12\u001b[0m%|█▏        |   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m6\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:29<09:56, \u001b[1m13.\u001b[0m55s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '8.66e-07', 'grad_norm': '1.851', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '9e-09', 'num_tokens': '4.018e+05', 'completions/mean_length':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '28.62', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '44', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '28.62',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '44',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8516',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.3569',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9136',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2828', 'reward': '1.308',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.4707', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '6.785e-06', 'entropy': '0.01261', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '12.11', 'epoch': '0.048'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:04:20] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             12%|█▏        | 6/50 [01:29<09:56, 13.55s/it]#015 14%|█▍        |                     \n",
-       "                             7/50 [01:41<09:24, 13.13s/it]#015                                                     \n",
-       "                             #015{'loss': '-0.0003451', 'grad_norm': '5.135', 'learning_rate':                     \n",
-       "                             '8.8e-09', 'num_tokens': '4.587e+05', 'completions/mean_length':                      \n",
-       "                             '27.24', 'completions/min_length': '21', 'completions/max_length':                    \n",
-       "                             '44', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '27.24',                                        \n",
-       "                             'completions/min_terminated_length': '21',                                            \n",
-       "                             'completions/max_terminated_length': '44',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7344',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4434',                                             \n",
-       "                             'rewards/format_reward/mean': '0.8822',                                               \n",
-       "                             'rewards/format_reward/std': '0.3246', 'reward': '1.175',                             \n",
-       "                             'reward_std': '0.5569', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.0001488', 'entropy': '0.01623', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '12.26', 'epoch': '0.056'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:04:20]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029259;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029260;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m12\u001b[0m%|█▏        | \u001b[1m6\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:29<09:56, \u001b[1m13.\u001b[0m55s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m14\u001b[0m%|█▍        |   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m7\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:41<09:24, \u001b[1m13.\u001b[0m13s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.0003451', 'grad_norm': '5.135', 'learning_rate':   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '8.8e-09', 'num_tokens': '4.587e+05', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '27.24', 'completions/min_length': '21', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '44', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '27.24',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '21',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '44',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7344',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4434',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8822',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.3246', 'reward': '1.175',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5569', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0001488', 'entropy': '0.01623', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '12.26', 'epoch': '0.056'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:04:31] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             14%|█▍        | 7/50 [01:41<09:24, 13.13s/it]#015 16%|█▌        |                     \n",
-       "                             8/50 [01:56<09:34, 13.69s/it]#015                                                     \n",
-       "                             #015{'loss': '0.006165', 'grad_norm': '1.91', 'learning_rate':                        \n",
-       "                             '8.6e-09', 'num_tokens': '5.259e+05', 'completions/mean_length':                      \n",
-       "                             '38.94', 'completions/min_length': '20', 'completions/max_length':                    \n",
-       "                             '109', 'completions/clipped_ratio': '0',                                              \n",
-       "                             'completions/mean_terminated_length': '38.94',                                        \n",
-       "                             'completions/min_terminated_length': '20',                                            \n",
-       "                             'completions/max_terminated_length': '109',                                           \n",
-       "                             'rewards/tool_call_reward/mean': '0.6797',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4684',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9685',                                               \n",
-       "                             'rewards/format_reward/std': '0.176', 'reward': '1.164',                              \n",
-       "                             'reward_std': '0.4987', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '5.315e-05', 'entropy': '0.01854', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '14.87', 'epoch': '0.064'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:04:31]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029267;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029268;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m14\u001b[0m%|█▍        | \u001b[1m7\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:41<09:24, \u001b[1m13.\u001b[0m13s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m16\u001b[0m%|█▌        |   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m8\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:56<09:34, \u001b[1m13.\u001b[0m69s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.006165', 'grad_norm': '1.91', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '8.6e-09', 'num_tokens': '5.259e+05', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '38.94', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '109', 'completions/clipped_ratio': '0',                            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '38.94',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '109',                         \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.6797',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4684',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9685',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.176', 'reward': '1.164',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.4987', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '5.315e-05', 'entropy': '0.01854', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '14.87', 'epoch': '0.064'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:04:47] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             16%|█▌        | 8/50 [01:56<09:34, 13.69s/it]#015 18%|█▊        |                     \n",
-       "                             9/50 [02:08<08:57, 13.10s/it]#015                                                     \n",
-       "                             #015{'loss': '0.001617', 'grad_norm': '2.566', 'learning_rate':                       \n",
-       "                             '8.4e-09', 'num_tokens': '5.847e+05', 'completions/mean_length':                      \n",
-       "                             '32.96', 'completions/min_length': '24', 'completions/max_length':                    \n",
-       "                             '51', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '32.96',                                        \n",
-       "                             'completions/min_terminated_length': '24',                                            \n",
-       "                             'completions/max_terminated_length': '51',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7969',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4039',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9215',                                               \n",
-       "                             'rewards/format_reward/std': '0.2708', 'reward': '1.258',                             \n",
-       "                             'reward_std': '0.4945', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '-1.58e-05', 'entropy': '0.01142', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '11.8', 'epoch': '0.072'}                                                \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:04:47]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029275;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029276;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m16\u001b[0m%|█▌        | \u001b[1m8\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m01:56<09:34, \u001b[1m13.\u001b[0m69s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m18\u001b[0m%|█▊        |   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m9\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:08<08:57, \u001b[1m13.\u001b[0m10s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.001617', 'grad_norm': '2.566', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '8.4e-09', 'num_tokens': '5.847e+05', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '32.96', 'completions/min_length': '24', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '51', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '32.96',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '24',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '51',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7969',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4039',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9215',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2708', 'reward': '1.258',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.4945', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '-1.58e-05', 'entropy': '0.01142', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '11.8', 'epoch': '0.072'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:04:58] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             18%|█▊        | 9/50 [02:08<08:57, 13.10s/it]#015 20%|██        |                     \n",
-       "                             10/50 [02:22<09:01, 13.54s/it]#015                                                    \n",
-       "                             #015{'loss': '-0.003781', 'grad_norm': '3.136', 'learning_rate':                      \n",
-       "                             '8.2e-09', 'num_tokens': '6.407e+05', 'completions/mean_length':                      \n",
-       "                             '30.45', 'completions/min_length': '19', 'completions/max_length':                    \n",
-       "                             '56', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '30.45',                                        \n",
-       "                             'completions/min_terminated_length': '19',                                            \n",
-       "                             'completions/max_terminated_length': '56',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7188',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4514',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9293',                                               \n",
-       "                             'rewards/format_reward/std': '0.2581', 'reward': '1.183',                             \n",
-       "                             'reward_std': '0.5212', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '5.048e-05', 'entropy': '0.01439', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '14.52', 'epoch': '0.08'}                                                \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:04:58]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029283;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029284;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m18\u001b[0m%|█▊        | \u001b[1m9\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:08<08:57, \u001b[1m13.\u001b[0m10s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m20\u001b[0m%|██        |   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m10\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:22<09:01, \u001b[1m13.\u001b[0m54s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.003781', 'grad_norm': '3.136', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '8.2e-09', 'num_tokens': '6.407e+05', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '30.45', 'completions/min_length': '19', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '56', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '30.45',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '19',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '56',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7188',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4514',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9293',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2581', 'reward': '1.183',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5212', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '5.048e-05', 'entropy': '0.01439', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '14.52', 'epoch': '0.08'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:05:14] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             20%|██        | 10/50 [02:22<09:01, 13.54s/it]#015 22%|██▏       |                    \n",
-       "                             11/50 [02:35<08:38, 13.31s/it]#015                                                    \n",
-       "                             #015{'loss': '0.00183', 'grad_norm': '1.218', 'learning_rate':                        \n",
-       "                             '8e-09', 'num_tokens': '6.995e+05', 'completions/mean_length':                        \n",
-       "                             '31.61', 'completions/min_length': '23', 'completions/max_length':                    \n",
-       "                             '47', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '31.61',                                        \n",
-       "                             'completions/min_terminated_length': '23',                                            \n",
-       "                             'completions/max_terminated_length': '47',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.8047',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.398',                                              \n",
-       "                             'rewards/format_reward/mean': '0.9372',                                               \n",
-       "                             'rewards/format_reward/std': '0.2442', 'reward': '1.273',                             \n",
-       "                             'reward_std': '0.4736', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '3.313e-05', 'entropy': '0.01158', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '12.76', 'epoch': '0.088'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:05:14]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029291;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029292;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m20\u001b[0m%|██        | \u001b[1m10\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:22<09:01, \u001b[1m13.\u001b[0m54s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m22\u001b[0m%|██▏       |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m11\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:35<08:38, \u001b[1m13.\u001b[0m31s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.00183', 'grad_norm': '1.218', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '8e-09', 'num_tokens': '6.995e+05', 'completions/mean_length':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '31.61', 'completions/min_length': '23', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '47', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '31.61',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '23',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '47',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8047',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.398',                            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9372',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2442', 'reward': '1.273',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.4736', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '3.313e-05', 'entropy': '0.01158', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '12.76', 'epoch': '0.088'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:05:26] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             22%|██▏       | 11/50 [02:35<08:38, 13.31s/it]#015 24%|██▍       |                    \n",
-       "                             12/50 [02:47<08:16, 13.07s/it]#015                                                    \n",
-       "                             #015{'loss': '-0.002743', 'grad_norm': '2.056', 'learning_rate':                      \n",
-       "                             '7.8e-09', 'num_tokens': '7.648e+05', 'completions/mean_length':                      \n",
-       "                             '31.52', 'completions/min_length': '22', 'completions/max_length':                    \n",
-       "                             '40', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '31.52',                                        \n",
-       "                             'completions/min_terminated_length': '22',                                            \n",
-       "                             'completions/max_terminated_length': '40',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7344',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4434',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9136',                                               \n",
-       "                             'rewards/format_reward/std': '0.283', 'reward': '1.191',                              \n",
-       "                             'reward_std': '0.5297', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '6.099e-05', 'entropy': '0.02272', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '12.53', 'epoch': '0.096'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:05:26]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029299;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029300;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m22\u001b[0m%|██▏       | \u001b[1m11\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:35<08:38, \u001b[1m13.\u001b[0m31s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m24\u001b[0m%|██▍       |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m12\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:47<08:16, \u001b[1m13.\u001b[0m07s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.002743', 'grad_norm': '2.056', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '7.8e-09', 'num_tokens': '7.648e+05', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '31.52', 'completions/min_length': '22', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '40', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '31.52',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '22',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '40',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7344',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4434',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9136',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.283', 'reward': '1.191',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5297', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '6.099e-05', 'entropy': '0.02272', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '12.53', 'epoch': '0.096'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:05:42] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             24%|██▍       | 12/50 [02:47<08:16, 13.07s/it]#015 26%|██▌       |                    \n",
-       "                             13/50 [03:05<08:48, 14.29s/it]#015                                                    \n",
-       "                             #015{'loss': '0.001483', 'grad_norm': '1.707', 'learning_rate':                       \n",
-       "                             '7.6e-09', 'num_tokens': '8.236e+05', 'completions/mean_length':                      \n",
-       "                             '32.16', 'completions/min_length': '21', 'completions/max_length':                    \n",
-       "                             '59', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '32.16',                                        \n",
-       "                             'completions/min_terminated_length': '21',                                            \n",
-       "                             'completions/max_terminated_length': '59',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7109',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4551',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9607',                                               \n",
-       "                             'rewards/format_reward/std': '0.1956', 'reward': '1.191',                             \n",
-       "                             'reward_std': '0.4948', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.0001074', 'entropy': '0.01953', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '17.07', 'epoch': '0.104'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:05:42]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029307;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029308;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m24\u001b[0m%|██▍       | \u001b[1m12\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m02:47<08:16, \u001b[1m13.\u001b[0m07s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m26\u001b[0m%|██▌       |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m13\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:05<08:48, \u001b[1m14.\u001b[0m29s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.001483', 'grad_norm': '1.707', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '7.6e-09', 'num_tokens': '8.236e+05', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '32.16', 'completions/min_length': '21', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '59', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '32.16',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '21',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '59',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7109',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4551',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9607',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.1956', 'reward': '1.191',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.4948', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0001074', 'entropy': '0.01953', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '17.07', 'epoch': '0.104'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:05:53] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             26%|██▌       | 13/50 [03:05<08:48, 14.29s/it]#015 28%|██▊       |                    \n",
-       "                             14/50 [03:19<08:32, 14.24s/it]#015                                                    \n",
-       "                             #015{'loss': '0.02113', 'grad_norm': '5.839', 'learning_rate':                        \n",
-       "                             '7.4e-09', 'num_tokens': '8.798e+05', 'completions/mean_length':                      \n",
-       "                             '35.77', 'completions/min_length': '15', 'completions/max_length':                    \n",
-       "                             '73', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '35.77',                                        \n",
-       "                             'completions/min_terminated_length': '15',                                            \n",
-       "                             'completions/max_terminated_length': '73',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.8047',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.398',                                              \n",
-       "                             'rewards/format_reward/mean': '0.9448',                                               \n",
-       "                             'rewards/format_reward/std': '0.2302', 'reward': '1.277',                             \n",
-       "                             'reward_std': '0.4652', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '9.532e-05', 'entropy': '0.02313', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '14.13', 'epoch': '0.112'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:05:53]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029315;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029316;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m26\u001b[0m%|██▌       | \u001b[1m13\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:05<08:48, \u001b[1m14.\u001b[0m29s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m28\u001b[0m%|██▊       |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m14\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:19<08:32, \u001b[1m14.\u001b[0m24s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.02113', 'grad_norm': '5.839', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '7.4e-09', 'num_tokens': '8.798e+05', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '35.77', 'completions/min_length': '15', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '73', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '35.77',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '15',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '73',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8047',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.398',                            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9448',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2302', 'reward': '1.277',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.4652', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '9.532e-05', 'entropy': '0.02313', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '14.13', 'epoch': '0.112'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:06:15] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             28%|██▊       | 14/50 [03:19<08:32, 14.24s/it]#015 30%|███       |                    \n",
-       "                             15/50 [03:36<08:51, 15.17s/it]#015                                                    \n",
-       "                             #015{'loss': '0.001838', 'grad_norm': '1.245', 'learning_rate':                       \n",
-       "                             '7.2e-09', 'num_tokens': '9.451e+05', 'completions/mean_length':                      \n",
-       "                             '42.38', 'completions/min_length': '20', 'completions/max_length':                    \n",
-       "                             '128', 'completions/clipped_ratio': '0.0625',                                         \n",
-       "                             'completions/mean_terminated_length': '36.67',                                        \n",
-       "                             'completions/min_terminated_length': '20',                                            \n",
-       "                             'completions/max_terminated_length': '62',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.8359',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.3718',                                             \n",
-       "                             'rewards/format_reward/mean': '0.8427',                                               \n",
-       "                             'rewards/format_reward/std': '0.367', 'reward': '1.257',                              \n",
-       "                             'reward_std': '0.5518', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '2.964e-05', 'entropy': '0.009179', 'clip_ratio/low_mean': '0',                       \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '17.31', 'epoch': '0.12'}                                                \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:06:15]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029323;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029324;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m28\u001b[0m%|██▊       | \u001b[1m14\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:19<08:32, \u001b[1m14.\u001b[0m24s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m30\u001b[0m%|███       |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m15\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:36<08:51, \u001b[1m15.\u001b[0m17s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.001838', 'grad_norm': '1.245', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '7.2e-09', 'num_tokens': '9.451e+05', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '42.38', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '128', 'completions/clipped_ratio': '0.0625',                       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '36.67',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '62',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8359',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.3718',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8427',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.367', 'reward': '1.257',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5518', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '2.964e-05', 'entropy': '0.009179', 'clip_ratio/low_mean': '0',     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '17.31', 'epoch': '0.12'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:06:31] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             30%|███       | 15/50 [03:36<08:51, 15.17s/it]#015 32%|███▏      |                    \n",
-       "                             16/50 [03:51<08:38, 15.25s/it]#015                                                    \n",
-       "                             #015{'loss': '-0.001349', 'grad_norm': '1.989', 'learning_rate':                      \n",
-       "                             '7e-09', 'num_tokens': '1.022e+06', 'completions/mean_length':                        \n",
-       "                             '32.54', 'completions/min_length': '16', 'completions/max_length':                    \n",
-       "                             '57', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '32.54',                                        \n",
-       "                             'completions/min_terminated_length': '16',                                            \n",
-       "                             'completions/max_terminated_length': '57',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7734',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4203',                                             \n",
-       "                             'rewards/format_reward/mean': '0.8978',                                               \n",
-       "                             'rewards/format_reward/std': '0.305', 'reward': '1.222',                              \n",
-       "                             'reward_std': '0.5287', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '8.802e-05', 'entropy': '0.02825', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '15.43', 'epoch': '0.128'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:06:31]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029331;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029332;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m30\u001b[0m%|███       | \u001b[1m15\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:36<08:51, \u001b[1m15.\u001b[0m17s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m32\u001b[0m%|███▏      |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m16\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:51<08:38, \u001b[1m15.\u001b[0m25s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.001349', 'grad_norm': '1.989', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '7e-09', 'num_tokens': '1.022e+06', 'completions/mean_length':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '32.54', 'completions/min_length': '16', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '57', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '32.54',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '16',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '57',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7734',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4203',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8978',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.305', 'reward': '1.222',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5287', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '8.802e-05', 'entropy': '0.02825', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '15.43', 'epoch': '0.128'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:06:41] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             32%|███▏      | 16/50 [03:51<08:38, 15.25s/it]#015 34%|███▍      |                    \n",
-       "                             17/50 [04:07<08:27, 15.38s/it]#015                                                    \n",
-       "                             #015{'loss': '0.0009151', 'grad_norm': '1.993', 'learning_rate':                      \n",
-       "                             '6.8e-09', 'num_tokens': '1.075e+06', 'completions/mean_length':                      \n",
-       "                             '35.89', 'completions/min_length': '20', 'completions/max_length':                    \n",
-       "                             '128', 'completions/clipped_ratio': '0.0625',                                         \n",
-       "                             'completions/mean_terminated_length': '29.75',                                        \n",
-       "                             'completions/min_terminated_length': '20',                                            \n",
-       "                             'completions/max_terminated_length': '45',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.8516',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.3569',                                             \n",
-       "                             'rewards/format_reward/mean': '0.8663',                                               \n",
-       "                             'rewards/format_reward/std': '0.3431', 'reward': '1.285',                             \n",
-       "                             'reward_std': '0.5211', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '1.607e-05', 'entropy': '0.009799', 'clip_ratio/low_mean': '0',                       \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '15.67', 'epoch': '0.136'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:06:41]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029339;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029340;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m32\u001b[0m%|███▏      | \u001b[1m16\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m03:51<08:38, \u001b[1m15.\u001b[0m25s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m34\u001b[0m%|███▍      |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m17\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:07<08:27, \u001b[1m15.\u001b[0m38s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.0009151', 'grad_norm': '1.993', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '6.8e-09', 'num_tokens': '1.075e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '35.89', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '128', 'completions/clipped_ratio': '0.0625',                       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '29.75',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '45',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8516',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.3569',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8663',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.3431', 'reward': '1.285',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5211', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '1.607e-05', 'entropy': '0.009799', 'clip_ratio/low_mean': '0',     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '15.67', 'epoch': '0.136'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:06:57] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             34%|███▍      | 17/50 [04:07<08:27, 15.38s/it]#015 36%|███▌      |                    \n",
-       "                             18/50 [04:20<07:46, 14.57s/it]#015                                                    \n",
-       "                             #015{'loss': '0.00337', 'grad_norm': '1.611', 'learning_rate':                        \n",
-       "                             '6.6e-09', 'num_tokens': '1.14e+06', 'completions/mean_length':                       \n",
-       "                             '33.16', 'completions/min_length': '20', 'completions/max_length':                    \n",
-       "                             '65', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '33.16',                                        \n",
-       "                             'completions/min_terminated_length': '20',                                            \n",
-       "                             'completions/max_terminated_length': '65',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7656',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4253',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9293',                                               \n",
-       "                             'rewards/format_reward/std': '0.2581', 'reward': '1.23',                              \n",
-       "                             'reward_std': '0.5021', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '5.279e-05', 'entropy': '0.02029', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '12.65', 'epoch': '0.144'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:06:57]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029347;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029348;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m34\u001b[0m%|███▍      | \u001b[1m17\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:07<08:27, \u001b[1m15.\u001b[0m38s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m36\u001b[0m%|███▌      |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m18\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:20<07:46, \u001b[1m14.\u001b[0m57s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.00337', 'grad_norm': '1.611', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '6.6e-09', 'num_tokens': '1.14e+06', 'completions/mean_length':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '33.16', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '65', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '33.16',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '65',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7656',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4253',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9293',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2581', 'reward': '1.23',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5021', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '5.279e-05', 'entropy': '0.02029', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '12.65', 'epoch': '0.144'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:07:08] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             36%|███▌      | 18/50 [04:20<07:46, 14.57s/it]#015 38%|███▊      |                    \n",
-       "                             19/50 [04:34<07:29, 14.51s/it]#015                                                    \n",
-       "                             #015{'loss': '-0.00885', 'grad_norm': '2.367', 'learning_rate':                       \n",
-       "                             '6.4e-09', 'num_tokens': '1.201e+06', 'completions/mean_length':                      \n",
-       "                             '33.41', 'completions/min_length': '16', 'completions/max_length':                    \n",
-       "                             '50', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '33.41',                                        \n",
-       "                             'completions/min_terminated_length': '16',                                            \n",
-       "                             'completions/max_terminated_length': '50',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.8672',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.3407',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9686',                                               \n",
-       "                             'rewards/format_reward/std': '0.1757', 'reward': '1.351',                             \n",
-       "                             'reward_std': '0.3889', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.0001576', 'entropy': '0.01414', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '14.36', 'epoch': '0.152'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:07:08]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029355;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029356;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m36\u001b[0m%|███▌      | \u001b[1m18\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:20<07:46, \u001b[1m14.\u001b[0m57s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m38\u001b[0m%|███▊      |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m19\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:34<07:29, \u001b[1m14.\u001b[0m51s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.00885', 'grad_norm': '2.367', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '6.4e-09', 'num_tokens': '1.201e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '33.41', 'completions/min_length': '16', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '50', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '33.41',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '16',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '50',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8672',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.3407',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9686',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.1757', 'reward': '1.351',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.3889', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0001576', 'entropy': '0.01414', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '14.36', 'epoch': '0.152'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:07:24] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             38%|███▊      | 19/50 [04:34<07:29, 14.51s/it]#015 40%|████      |                    \n",
-       "                             20/50 [04:46<06:53, 13.79s/it]#015                                                    \n",
-       "                             #015{'loss': '-0.003722', 'grad_norm': '2.091', 'learning_rate':                      \n",
-       "                             '6.2e-09', 'num_tokens': '1.272e+06', 'completions/mean_length':                      \n",
-       "                             '32.25', 'completions/min_length': '19', 'completions/max_length':                    \n",
-       "                             '53', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '32.25',                                        \n",
-       "                             'completions/min_terminated_length': '19',                                            \n",
-       "                             'completions/max_terminated_length': '53',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7344',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4434',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9293',                                               \n",
-       "                             'rewards/format_reward/std': '0.258', 'reward': '1.199',                              \n",
-       "                             'reward_std': '0.5153', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '8.643e-05', 'entropy': '0.03131', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '12.09', 'epoch': '0.16'}                                                \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:07:24]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029363;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029364;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m38\u001b[0m%|███▊      | \u001b[1m19\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:34<07:29, \u001b[1m14.\u001b[0m51s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m40\u001b[0m%|████      |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m20\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:46<06:53, \u001b[1m13.\u001b[0m79s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.003722', 'grad_norm': '2.091', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '6.2e-09', 'num_tokens': '1.272e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '32.25', 'completions/min_length': '19', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '53', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '32.25',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '19',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '53',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7344',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4434',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9293',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.258', 'reward': '1.199',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5153', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '8.643e-05', 'entropy': '0.03131', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '12.09', 'epoch': '0.16'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:07:35] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             40%|████      | 20/50 [04:46<06:53, 13.79s/it]#015 42%|████▏     |                    \n",
-       "                             21/50 [04:58<06:24, 13.25s/it]#015                                                    \n",
-       "                             #015{'loss': '-0.0004618', 'grad_norm': '2.553', 'learning_rate':                     \n",
-       "                             '6e-09', 'num_tokens': '1.329e+06', 'completions/mean_length':                        \n",
-       "                             '30.9', 'completions/min_length': '20', 'completions/max_length':                     \n",
-       "                             '54', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '30.9',                                         \n",
-       "                             'completions/min_terminated_length': '20',                                            \n",
-       "                             'completions/max_terminated_length': '54',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7891',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4096',                                             \n",
-       "                             'rewards/format_reward/mean': '0.8901',                                               \n",
-       "                             'rewards/format_reward/std': '0.315', 'reward': '1.234',                              \n",
-       "                             'reward_std': '0.5291', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.0001838', 'entropy': '0.01313', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '11.98', 'epoch': '0.168'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:07:35]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029371;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029372;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m40\u001b[0m%|████      | \u001b[1m20\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:46<06:53, \u001b[1m13.\u001b[0m79s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m42\u001b[0m%|████▏     |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m21\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:58<06:24, \u001b[1m13.\u001b[0m25s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.0004618', 'grad_norm': '2.553', 'learning_rate':   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '6e-09', 'num_tokens': '1.329e+06', 'completions/mean_length':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '30.9', 'completions/min_length': '20', 'completions/max_length':   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '54', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '30.9',                       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '54',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7891',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4096',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8901',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.315', 'reward': '1.234',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5291', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0001838', 'entropy': '0.01313', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '11.98', 'epoch': '0.168'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:07:51] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             42%|████▏     | 21/50 [04:58<06:24, 13.25s/it]#015 44%|████▍     |                    \n",
-       "                             22/50 [05:12<06:17, 13.48s/it]#015                                                    \n",
-       "                             #015{'loss': '0.003621', 'grad_norm': '2.035', 'learning_rate':                       \n",
-       "                             '5.8e-09', 'num_tokens': '1.393e+06', 'completions/mean_length':                      \n",
-       "                             '30.68', 'completions/min_length': '18', 'completions/max_length':                    \n",
-       "                             '45', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '30.68',                                        \n",
-       "                             'completions/min_terminated_length': '18',                                            \n",
-       "                             'completions/max_terminated_length': '45',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7031',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4587',                                             \n",
-       "                             'rewards/format_reward/mean': '0.945', 'rewards/format_reward/std':                   \n",
-       "                             '0.2295', 'reward': '1.176', 'reward_std': '0.5124',                                  \n",
-       "                             'frac_reward_zero_std': '0', 'kl': '0.0001385', 'entropy':                            \n",
-       "                             '0.01903', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',                     \n",
-       "                             'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',                              \n",
-       "                             'clip_ratio/region_mean': '0', 'step_time': '13.99', 'epoch':                         \n",
-       "                             '0.176'}                                                                              \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:07:51]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029379;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029380;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m42\u001b[0m%|████▏     | \u001b[1m21\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m04:58<06:24, \u001b[1m13.\u001b[0m25s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m44\u001b[0m%|████▍     |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m22\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:12<06:17, \u001b[1m13.\u001b[0m48s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.003621', 'grad_norm': '2.035', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '5.8e-09', 'num_tokens': '1.393e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '30.68', 'completions/min_length': '18', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '45', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '30.68',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '18',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '45',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7031',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4587',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.945', 'rewards/format_reward/std': \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.2295', 'reward': '1.176', 'reward_std': '0.5124',                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'frac_reward_zero_std': '0', 'kl': '0.0001385', 'entropy':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.01903', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/region_mean': '0', 'step_time': '13.99', 'epoch':       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.176'\u001b[1m}\u001b[0m                                                            \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:08:07] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             44%|████▍     | 22/50 [05:12<06:17, 13.48s/it]#015 46%|████▌     |                    \n",
-       "                             23/50 [05:28<06:22, 14.15s/it]#015                                                    \n",
-       "                             #015{'loss': '0.00291', 'grad_norm': '2.108', 'learning_rate':                        \n",
-       "                             '5.6e-09', 'num_tokens': '1.461e+06', 'completions/mean_length':                      \n",
-       "                             '38.35', 'completions/min_length': '17', 'completions/max_length':                    \n",
-       "                             '128', 'completions/clipped_ratio': '0.0625',                                         \n",
-       "                             'completions/mean_terminated_length': '32.38',                                        \n",
-       "                             'completions/min_terminated_length': '17',                                            \n",
-       "                             'completions/max_terminated_length': '47',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.8047',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.398',                                              \n",
-       "                             'rewards/format_reward/mean': '0.8349',                                               \n",
-       "                             'rewards/format_reward/std': '0.3741', 'reward': '1.222',                             \n",
-       "                             'reward_std': '0.5721', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.0001697', 'entropy': '0.01636', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '15.71', 'epoch': '0.184'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:08:07]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029387;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029388;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m44\u001b[0m%|████▍     | \u001b[1m22\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:12<06:17, \u001b[1m13.\u001b[0m48s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m46\u001b[0m%|████▌     |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m23\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:28<06:22, \u001b[1m14.\u001b[0m15s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.00291', 'grad_norm': '2.108', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '5.6e-09', 'num_tokens': '1.461e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '38.35', 'completions/min_length': '17', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '128', 'completions/clipped_ratio': '0.0625',                       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '32.38',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '17',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '47',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8047',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.398',                            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8349',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.3741', 'reward': '1.222',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5721', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0001697', 'entropy': '0.01636', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '15.71', 'epoch': '0.184'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:08:18] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             46%|████▌     | 23/50 [05:28<06:22, 14.15s/it]#015 48%|████▊     |                    \n",
-       "                             24/50 [05:41<06:02, 13.93s/it]#015                                                    \n",
-       "                             #015{'loss': '0.001389', 'grad_norm': '1.385', 'learning_rate':                       \n",
-       "                             '5.4e-09', 'num_tokens': '1.526e+06', 'completions/mean_length':                      \n",
-       "                             '39.6', 'completions/min_length': '21', 'completions/max_length':                     \n",
-       "                             '84', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '39.6',                                         \n",
-       "                             'completions/min_terminated_length': '21',                                            \n",
-       "                             'completions/max_terminated_length': '84',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.5625',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.498',                                              \n",
-       "                             'rewards/format_reward/mean': '0.9135',                                               \n",
-       "                             'rewards/format_reward/std': '0.2831', 'reward': '1.019',                             \n",
-       "                             'reward_std': '0.5631', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.0001944', 'entropy': '0.06599', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '13.4', 'epoch': '0.192'}                                                \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:08:18]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029395;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029396;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m46\u001b[0m%|████▌     | \u001b[1m23\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:28<06:22, \u001b[1m14.\u001b[0m15s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m48\u001b[0m%|████▊     |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m24\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:41<06:02, \u001b[1m13.\u001b[0m93s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.001389', 'grad_norm': '1.385', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '5.4e-09', 'num_tokens': '1.526e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '39.6', 'completions/min_length': '21', 'completions/max_length':   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '84', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '39.6',                       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '21',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '84',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.5625',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.498',                            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9135',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2831', 'reward': '1.019',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5631', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0001944', 'entropy': '0.06599', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '13.4', 'epoch': '0.192'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:08:34] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             48%|████▊     | 24/50 [05:41<06:02, 13.93s/it]#015 50%|█████     |                    \n",
-       "                             25/50 [05:56<05:54, 14.18s/it]#015                                                    \n",
-       "                             #015{'loss': '0.01243', 'grad_norm': '4.469', 'learning_rate':                        \n",
-       "                             '5.2e-09', 'num_tokens': '1.586e+06', 'completions/mean_length':                      \n",
-       "                             '36.21', 'completions/min_length': '19', 'completions/max_length':                    \n",
-       "                             '56', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '36.21',                                        \n",
-       "                             'completions/min_terminated_length': '19',                                            \n",
-       "                             'completions/max_terminated_length': '56',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.8203',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.3854',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9372',                                               \n",
-       "                             'rewards/format_reward/std': '0.2443', 'reward': '1.289',                             \n",
-       "                             'reward_std': '0.4641', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.0001467', 'entropy': '0.01645', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '14.75', 'epoch': '0.2'}                                                 \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:08:34]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029403;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029404;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m48\u001b[0m%|████▊     | \u001b[1m24\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:41<06:02, \u001b[1m13.\u001b[0m93s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m50\u001b[0m%|█████     |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m25\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:56<05:54, \u001b[1m14.\u001b[0m18s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.01243', 'grad_norm': '4.469', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '5.2e-09', 'num_tokens': '1.586e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '36.21', 'completions/min_length': '19', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '56', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '36.21',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '19',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '56',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8203',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.3854',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9372',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2443', 'reward': '1.289',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.4641', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0001467', 'entropy': '0.01645', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '14.75', 'epoch': '0.2'\u001b[1m}\u001b[0m                               \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:08:50] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             50%|█████     | 25/50 [05:56<05:54, 14.18s/it]#015 52%|█████▏    |                    \n",
-       "                             26/50 [06:12<05:51, 14.63s/it]#015                                                    \n",
-       "                             #015{'loss': '0.01857', 'grad_norm': '4.155', 'learning_rate':                        \n",
-       "                             '5e-09', 'num_tokens': '1.65e+06', 'completions/mean_length':                         \n",
-       "                             '32.46', 'completions/min_length': '21', 'completions/max_length':                    \n",
-       "                             '128', 'completions/clipped_ratio': '0.007812',                                       \n",
-       "                             'completions/mean_terminated_length': '31.71',                                        \n",
-       "                             'completions/min_terminated_length': '21',                                            \n",
-       "                             'completions/max_terminated_length': '66',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7344',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4434',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9369',                                               \n",
-       "                             'rewards/format_reward/std': '0.2452', 'reward': '1.203',                             \n",
-       "                             'reward_std': '0.5082', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.0001655', 'entropy': '0.0276', 'clip_ratio/low_mean': '0',                         \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '15.67', 'epoch': '0.208'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:08:50]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029411;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029412;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m50\u001b[0m%|█████     | \u001b[1m25\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m05:56<05:54, \u001b[1m14.\u001b[0m18s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m52\u001b[0m%|█████▏    |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m26\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:12<05:51, \u001b[1m14.\u001b[0m63s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.01857', 'grad_norm': '4.155', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '5e-09', 'num_tokens': '1.65e+06', 'completions/mean_length':       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '32.46', 'completions/min_length': '21', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '128', 'completions/clipped_ratio': '0.007812',                     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '31.71',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '21',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '66',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7344',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4434',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9369',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2452', 'reward': '1.203',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5082', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0001655', 'entropy': '0.0276', 'clip_ratio/low_mean': '0',       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '15.67', 'epoch': '0.208'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:09:07] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             52%|█████▏    | 26/50 [06:12<05:51, 14.63s/it]#015 54%|█████▍    |                    \n",
-       "                             27/50 [06:27<05:43, 14.92s/it]#015                                                    \n",
-       "                             #015{'loss': '0.03352', 'grad_norm': '1.645', 'learning_rate':                        \n",
-       "                             '4.8e-09', 'num_tokens': '1.719e+06', 'completions/mean_length':                      \n",
-       "                             '39.02', 'completions/min_length': '20', 'completions/max_length':                    \n",
-       "                             '128', 'completions/clipped_ratio': '0.05469',                                        \n",
-       "                             'completions/mean_terminated_length': '33.88',                                        \n",
-       "                             'completions/min_terminated_length': '20',                                            \n",
-       "                             'completions/max_terminated_length': '59',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.6953',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4621',                                             \n",
-       "                             'rewards/format_reward/mean': '0.874', 'rewards/format_reward/std':                   \n",
-       "                             '0.3347', 'reward': '1.132', 'reward_std': '0.5743',                                  \n",
-       "                             'frac_reward_zero_std': '0', 'kl': '2.56e-05', 'entropy':                             \n",
-       "                             '0.01199', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',                     \n",
-       "                             'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',                              \n",
-       "                             'clip_ratio/region_mean': '0', 'step_time': '15.6', 'epoch':                          \n",
-       "                             '0.216'}                                                                              \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:09:07]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029419;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029420;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m52\u001b[0m%|█████▏    | \u001b[1m26\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:12<05:51, \u001b[1m14.\u001b[0m63s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m54\u001b[0m%|█████▍    |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m27\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:27<05:43, \u001b[1m14.\u001b[0m92s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.03352', 'grad_norm': '1.645', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '4.8e-09', 'num_tokens': '1.719e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '39.02', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '128', 'completions/clipped_ratio': '0.05469',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '33.88',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '59',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.6953',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4621',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.874', 'rewards/format_reward/std': \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.3347', 'reward': '1.132', 'reward_std': '0.5743',                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'frac_reward_zero_std': '0', 'kl': '2.56e-05', 'entropy':           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.01199', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/region_mean': '0', 'step_time': '15.6', 'epoch':        \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.216'\u001b[1m}\u001b[0m                                                            \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:09:18] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             54%|█████▍    | 27/50 [06:27<05:43, 14.92s/it]#015 56%|█████▌    |                    \n",
-       "                             28/50 [06:40<05:13, 14.26s/it]#015                                                    \n",
-       "                             #015{'loss': '-0.01125', 'grad_norm': '2.16', 'learning_rate':                        \n",
-       "                             '4.6e-09', 'num_tokens': '1.786e+06', 'completions/mean_length':                      \n",
-       "                             '33.49', 'completions/min_length': '19', 'completions/max_length':                    \n",
-       "                             '59', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '33.49',                                        \n",
-       "                             'completions/min_terminated_length': '19',                                            \n",
-       "                             'completions/max_terminated_length': '59',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.6797',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4684',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9214',                                               \n",
-       "                             'rewards/format_reward/std': '0.2709', 'reward': '1.14',                              \n",
-       "                             'reward_std': '0.54', 'frac_reward_zero_std': '0', 'kl':                              \n",
-       "                             '0.0001186', 'entropy': '0.01506', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '12.69', 'epoch': '0.224'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:09:18]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029427;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029428;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m54\u001b[0m%|█████▍    | \u001b[1m27\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:27<05:43, \u001b[1m14.\u001b[0m92s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m56\u001b[0m%|█████▌    |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m28\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:40<05:13, \u001b[1m14.\u001b[0m26s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.01125', 'grad_norm': '2.16', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '4.6e-09', 'num_tokens': '1.786e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '33.49', 'completions/min_length': '19', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '59', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '33.49',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '19',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '59',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.6797',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4684',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9214',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2709', 'reward': '1.14',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.54', 'frac_reward_zero_std': '0', 'kl':            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0001186', 'entropy': '0.01506', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '12.69', 'epoch': '0.224'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:09:29] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             56%|█████▌    | 28/50 [06:40<05:13, 14.26s/it]#015 58%|█████▊    |                    \n",
-       "                             29/50 [06:53<04:50, 13.82s/it]#015                                                    \n",
-       "                             #015{'loss': '0.00335', 'grad_norm': '1.61', 'learning_rate':                         \n",
-       "                             '4.4e-09', 'num_tokens': '1.85e+06', 'completions/mean_length':                       \n",
-       "                             '29.99', 'completions/min_length': '20', 'completions/max_length':                    \n",
-       "                             '46', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '29.99',                                        \n",
-       "                             'completions/min_terminated_length': '20',                                            \n",
-       "                             'completions/max_terminated_length': '46',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7969',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4039',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9372',                                               \n",
-       "                             'rewards/format_reward/std': '0.2442', 'reward': '1.265',                             \n",
-       "                             'reward_std': '0.478', 'frac_reward_zero_std': '0', 'kl':                             \n",
-       "                             '0.0001361', 'entropy': '0.03338', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '12.79', 'epoch': '0.232'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:09:29]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029435;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029436;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m56\u001b[0m%|█████▌    | \u001b[1m28\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:40<05:13, \u001b[1m14.\u001b[0m26s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m58\u001b[0m%|█████▊    |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m29\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:53<04:50, \u001b[1m13.\u001b[0m82s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.00335', 'grad_norm': '1.61', 'learning_rate':       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '4.4e-09', 'num_tokens': '1.85e+06', 'completions/mean_length':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '29.99', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '46', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '29.99',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '46',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7969',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4039',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9372',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2442', 'reward': '1.265',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.478', 'frac_reward_zero_std': '0', 'kl':           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0001361', 'entropy': '0.03338', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '12.79', 'epoch': '0.232'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:09:45] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             58%|█████▊    | 29/50 [06:53<04:50, 13.82s/it]#015 60%|██████    |                    \n",
-       "                             30/50 [07:09<04:49, 14.46s/it]#015                                                    \n",
-       "                             #015{'loss': '0.003962', 'grad_norm': '1.33', 'learning_rate':                        \n",
-       "                             '4.2e-09', 'num_tokens': '1.919e+06', 'completions/mean_length':                      \n",
-       "                             '42.09', 'completions/min_length': '19', 'completions/max_length':                    \n",
-       "                             '128', 'completions/clipped_ratio': '0.0625',                                         \n",
-       "                             'completions/mean_terminated_length': '36.36',                                        \n",
-       "                             'completions/min_terminated_length': '19',                                            \n",
-       "                             'completions/max_terminated_length': '90',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.6641',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4742',                                             \n",
-       "                             'rewards/format_reward/mean': '0.8181',                                               \n",
-       "                             'rewards/format_reward/std': '0.3903', 'reward': '1.073',                             \n",
-       "                             'reward_std': '0.6202', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.0001873', 'entropy': '0.04145', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '15.93', 'epoch': '0.24'}                                                \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:09:45]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029443;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029444;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m58\u001b[0m%|█████▊    | \u001b[1m29\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m06:53<04:50, \u001b[1m13.\u001b[0m82s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m60\u001b[0m%|██████    |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m30\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:09<04:49, \u001b[1m14.\u001b[0m46s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.003962', 'grad_norm': '1.33', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '4.2e-09', 'num_tokens': '1.919e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '42.09', 'completions/min_length': '19', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '128', 'completions/clipped_ratio': '0.0625',                       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '36.36',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '19',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '90',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.6641',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4742',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8181',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.3903', 'reward': '1.073',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.6202', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0001873', 'entropy': '0.04145', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '15.93', 'epoch': '0.24'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:09:55] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             60%|██████    | 30/50 [07:09<04:49, 14.46s/it]#015 62%|██████▏   |                    \n",
-       "                             31/50 [07:20<04:17, 13.56s/it]#015                                                    \n",
-       "                             #015{'loss': '-0.005677', 'grad_norm': '2.497', 'learning_rate':                      \n",
-       "                             '4e-09', 'num_tokens': '1.974e+06', 'completions/mean_length':                        \n",
-       "                             '32.5', 'completions/min_length': '21', 'completions/max_length':                     \n",
-       "                             '41', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '32.5',                                         \n",
-       "                             'completions/min_terminated_length': '21',                                            \n",
-       "                             'completions/max_terminated_length': '41',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7656',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4253',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9136',                                               \n",
-       "                             'rewards/format_reward/std': '0.2829', 'reward': '1.222',                             \n",
-       "                             'reward_std': '0.5172', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.0001616', 'entropy': '0.01579', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '11.46', 'epoch': '0.248'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:09:55]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029451;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029452;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m60\u001b[0m%|██████    | \u001b[1m30\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:09<04:49, \u001b[1m14.\u001b[0m46s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m62\u001b[0m%|██████▏   |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m31\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:20<04:17, \u001b[1m13.\u001b[0m56s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.005677', 'grad_norm': '2.497', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '4e-09', 'num_tokens': '1.974e+06', 'completions/mean_length':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '32.5', 'completions/min_length': '21', 'completions/max_length':   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '41', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '32.5',                       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '21',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '41',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7656',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4253',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9136',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2829', 'reward': '1.222',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5172', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0001616', 'entropy': '0.01579', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '11.46', 'epoch': '0.248'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:10:11] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             62%|██████▏   | 31/50 [07:20<04:17, 13.56s/it]#015 64%|██████▍   |                    \n",
-       "                             32/50 [07:36<04:13, 14.07s/it]#015                                                    \n",
-       "                             #015{'loss': '-0.0004137', 'grad_norm': '2.191', 'learning_rate':                     \n",
-       "                             '3.8e-09', 'num_tokens': '2.054e+06', 'completions/mean_length':                      \n",
-       "                             '35.31', 'completions/min_length': '22', 'completions/max_length':                    \n",
-       "                             '61', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '35.31',                                        \n",
-       "                             'completions/min_terminated_length': '22',                                            \n",
-       "                             'completions/max_terminated_length': '61',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7891',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4096',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9528',                                               \n",
-       "                             'rewards/format_reward/std': '0.2136', 'reward': '1.265',                             \n",
-       "                             'reward_std': '0.4655', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '4.369e-05', 'entropy': '0.0221', 'clip_ratio/low_mean': '0',                         \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '15.24', 'epoch': '0.256'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:10:11]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029459;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029460;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m62\u001b[0m%|██████▏   | \u001b[1m31\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:20<04:17, \u001b[1m13.\u001b[0m56s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m64\u001b[0m%|██████▍   |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m32\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:36<04:13, \u001b[1m14.\u001b[0m07s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.0004137', 'grad_norm': '2.191', 'learning_rate':   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '3.8e-09', 'num_tokens': '2.054e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '35.31', 'completions/min_length': '22', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '61', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '35.31',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '22',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '61',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7891',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4096',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9528',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2136', 'reward': '1.265',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.4655', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '4.369e-05', 'entropy': '0.0221', 'clip_ratio/low_mean': '0',       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '15.24', 'epoch': '0.256'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:10:27] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             64%|██████▍   | 32/50 [07:36<04:13, 14.07s/it]#015 66%|██████▌   |                    \n",
-       "                             33/50 [07:49<03:56, 13.92s/it]#015                                                    \n",
-       "                             #015{'loss': '0.007093', 'grad_norm': '2.214', 'learning_rate':                       \n",
-       "                             '3.6e-09', 'num_tokens': '2.112e+06', 'completions/mean_length':                      \n",
-       "                             '37.85', 'completions/min_length': '20', 'completions/max_length':                    \n",
-       "                             '84', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '37.85',                                        \n",
-       "                             'completions/min_terminated_length': '20',                                            \n",
-       "                             'completions/max_terminated_length': '84',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7734',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4203',                                             \n",
-       "                             'rewards/format_reward/mean': '0.945', 'rewards/format_reward/std':                   \n",
-       "                             '0.2294', 'reward': '1.246', 'reward_std': '0.4823',                                  \n",
-       "                             'frac_reward_zero_std': '0', 'kl': '0.0002727', 'entropy':                            \n",
-       "                             '0.02825', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',                     \n",
-       "                             'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',                              \n",
-       "                             'clip_ratio/region_mean': '0', 'step_time': '13.56', 'epoch':                         \n",
-       "                             '0.264'}                                                                              \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:10:27]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029467;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029468;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m64\u001b[0m%|██████▍   | \u001b[1m32\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:36<04:13, \u001b[1m14.\u001b[0m07s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m66\u001b[0m%|██████▌   |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m33\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:49<03:56, \u001b[1m13.\u001b[0m92s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.007093', 'grad_norm': '2.214', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '3.6e-09', 'num_tokens': '2.112e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '37.85', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '84', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '37.85',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '84',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7734',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4203',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.945', 'rewards/format_reward/std': \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.2294', 'reward': '1.246', 'reward_std': '0.4823',                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'frac_reward_zero_std': '0', 'kl': '0.0002727', 'entropy':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.02825', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/region_mean': '0', 'step_time': '13.56', 'epoch':       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.264'\u001b[1m}\u001b[0m                                                            \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:10:39] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             66%|██████▌   | 33/50 [07:49<03:56, 13.92s/it]#015 68%|██████▊   |                    \n",
-       "                             34/50 [08:02<03:35, 13.44s/it]#015                                                    \n",
-       "                             #015{'loss': '0.01603', 'grad_norm': '4.95', 'learning_rate':                         \n",
-       "                             '3.4e-09', 'num_tokens': '2.19e+06', 'completions/mean_length':                       \n",
-       "                             '29.18', 'completions/min_length': '20', 'completions/max_length':                    \n",
-       "                             '50', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '29.18',                                        \n",
-       "                             'completions/min_terminated_length': '20',                                            \n",
-       "                             'completions/max_terminated_length': '50',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.75',                                              \n",
-       "                             'rewards/tool_call_reward/std': '0.4347',                                             \n",
-       "                             'rewards/format_reward/mean': '0.8979',                                               \n",
-       "                             'rewards/format_reward/std': '0.3048', 'reward': '1.199',                             \n",
-       "                             'reward_std': '0.5379', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.000156', 'entropy': '0.0227', 'clip_ratio/low_mean': '0',                          \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '12.3', 'epoch': '0.272'}                                                \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:10:39]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029475;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029476;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m66\u001b[0m%|██████▌   | \u001b[1m33\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m07:49<03:56, \u001b[1m13.\u001b[0m92s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m68\u001b[0m%|██████▊   |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m34\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m08:02<03:35, \u001b[1m13.\u001b[0m44s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.01603', 'grad_norm': '4.95', 'learning_rate':       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '3.4e-09', 'num_tokens': '2.19e+06', 'completions/mean_length':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '29.18', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '50', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '29.18',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '50',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.75',                            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4347',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8979',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.3048', 'reward': '1.199',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5379', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.000156', 'entropy': '0.0227', 'clip_ratio/low_mean': '0',        \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '12.3', 'epoch': '0.272'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:10:55] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             68%|██████▊   | 34/50 [08:02<03:35, 13.44s/it]#015 70%|███████   |                    \n",
-       "                             35/50 [08:17<03:29, 13.96s/it]#015                                                    \n",
-       "                             #015{'loss': '0.002534', 'grad_norm': '1.978', 'learning_rate':                       \n",
-       "                             '3.2e-09', 'num_tokens': '2.246e+06', 'completions/mean_length':                      \n",
-       "                             '30.71', 'completions/min_length': '20', 'completions/max_length':                    \n",
-       "                             '57', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '30.71',                                        \n",
-       "                             'completions/min_terminated_length': '20',                                            \n",
-       "                             'completions/max_terminated_length': '57',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.8359',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.3718',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9058',                                               \n",
-       "                             'rewards/format_reward/std': '0.2941', 'reward': '1.289',                             \n",
-       "                             'reward_std': '0.4891', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '5.05e-05', 'entropy': '0.01136', 'clip_ratio/low_mean': '0',                         \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '15.18', 'epoch': '0.28'}                                                \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:10:55]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029483;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029484;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m68\u001b[0m%|██████▊   | \u001b[1m34\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m08:02<03:35, \u001b[1m13.\u001b[0m44s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m70\u001b[0m%|███████   |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m35\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m08:17<03:29, \u001b[1m13.\u001b[0m96s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.002534', 'grad_norm': '1.978', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '3.2e-09', 'num_tokens': '2.246e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '30.71', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '57', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '30.71',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '57',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8359',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.3718',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9058',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2941', 'reward': '1.289',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.4891', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '5.05e-05', 'entropy': '0.01136', 'clip_ratio/low_mean': '0',       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '15.18', 'epoch': '0.28'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:11:06] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             70%|███████   | 35/50 [08:17<03:29, 13.96s/it]#015 72%|███████▏  |                    \n",
-       "                             36/50 [08:32<03:20, 14.32s/it]#015                                                    \n",
-       "                             #015{'loss': '0.00398', 'grad_norm': '2.012', 'learning_rate':                        \n",
-       "                             '3e-09', 'num_tokens': '2.305e+06', 'completions/mean_length':                        \n",
-       "                             '34.98', 'completions/min_length': '19', 'completions/max_length':                    \n",
-       "                             '119', 'completions/clipped_ratio': '0',                                              \n",
-       "                             'completions/mean_terminated_length': '34.98',                                        \n",
-       "                             'completions/min_terminated_length': '19',                                            \n",
-       "                             'completions/max_terminated_length': '119',                                           \n",
-       "                             'rewards/tool_call_reward/mean': '0.6719',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4714',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9214',                                               \n",
-       "                             'rewards/format_reward/std': '0.2712', 'reward': '1.133',                             \n",
-       "                             'reward_std': '0.5421', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.0001751', 'entropy': '0.01811', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '15.15', 'epoch': '0.288'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:11:06]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029491;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029492;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m70\u001b[0m%|███████   | \u001b[1m35\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m08:17<03:29, \u001b[1m13.\u001b[0m96s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m72\u001b[0m%|███████▏  |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m36\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m08:32<03:20, \u001b[1m14.\u001b[0m32s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.00398', 'grad_norm': '2.012', 'learning_rate':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '3e-09', 'num_tokens': '2.305e+06', 'completions/mean_length':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '34.98', 'completions/min_length': '19', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '119', 'completions/clipped_ratio': '0',                            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '34.98',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '19',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '119',                         \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.6719',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4714',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9214',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2712', 'reward': '1.133',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5421', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0001751', 'entropy': '0.01811', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '15.15', 'epoch': '0.288'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:11:23] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             72%|███████▏  | 36/50 [08:32<03:20, 14.32s/it]#015 74%|███████▍  |                    \n",
-       "                             37/50 [08:46<03:04, 14.16s/it]#015                                                    \n",
-       "                             #015{'loss': '0.005717', 'grad_norm': '2.737', 'learning_rate':                       \n",
-       "                             '2.8e-09', 'num_tokens': '2.393e+06', 'completions/mean_length':                      \n",
-       "                             '31.75', 'completions/min_length': '22', 'completions/max_length':                    \n",
-       "                             '48', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '31.75',                                        \n",
-       "                             'completions/min_terminated_length': '22',                                            \n",
-       "                             'completions/max_terminated_length': '48',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7734',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4203',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9214',                                               \n",
-       "                             'rewards/format_reward/std': '0.2709', 'reward': '1.234',                             \n",
-       "                             'reward_std': '0.5062', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.0001182', 'entropy': '0.02041', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '13.78', 'epoch': '0.296'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:11:23]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029499;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029500;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m72\u001b[0m%|███████▏  | \u001b[1m36\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m08:32<03:20, \u001b[1m14.\u001b[0m32s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m74\u001b[0m%|███████▍  |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m37\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m08:46<03:04, \u001b[1m14.\u001b[0m16s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.005717', 'grad_norm': '2.737', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '2.8e-09', 'num_tokens': '2.393e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '31.75', 'completions/min_length': '22', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '48', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '31.75',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '22',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '48',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7734',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4203',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9214',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2709', 'reward': '1.234',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5062', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0001182', 'entropy': '0.02041', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '13.78', 'epoch': '0.296'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:11:39] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             74%|███████▍  | 37/50 [08:46<03:04, 14.16s/it]#015 76%|███████▌  |                    \n",
-       "                             38/50 [09:02<02:57, 14.76s/it]#015                                                    \n",
-       "                             #015{'loss': '-0.003074', 'grad_norm': '2.631', 'learning_rate':                      \n",
-       "                             '2.6e-09', 'num_tokens': '2.457e+06', 'completions/mean_length':                      \n",
-       "                             '33.31', 'completions/min_length': '21', 'completions/max_length':                    \n",
-       "                             '62', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '33.31',                                        \n",
-       "                             'completions/min_terminated_length': '21',                                            \n",
-       "                             'completions/max_terminated_length': '62',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7109',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4551',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9135',                                               \n",
-       "                             'rewards/format_reward/std': '0.2831', 'reward': '1.168',                             \n",
-       "                             'reward_std': '0.5377', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '4.227e-05', 'entropy': '0.0207', 'clip_ratio/low_mean': '0',                         \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '16.14', 'epoch': '0.304'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:11:39]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029507;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029508;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m74\u001b[0m%|███████▍  | \u001b[1m37\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m08:46<03:04, \u001b[1m14.\u001b[0m16s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m76\u001b[0m%|███████▌  |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m38\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:02<02:57, \u001b[1m14.\u001b[0m76s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.003074', 'grad_norm': '2.631', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '2.6e-09', 'num_tokens': '2.457e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '33.31', 'completions/min_length': '21', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '62', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '33.31',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '21',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '62',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7109',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4551',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9135',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2831', 'reward': '1.168',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5377', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '4.227e-05', 'entropy': '0.0207', 'clip_ratio/low_mean': '0',       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '16.14', 'epoch': '0.304'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:11:54] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             76%|███████▌  | 38/50 [09:02<02:57, 14.76s/it]#015 78%|███████▊  |                    \n",
-       "                             39/50 [09:18<02:46, 15.15s/it]#015                                                    \n",
-       "                             #015{'loss': '0.002493', 'grad_norm': '2.959', 'learning_rate':                       \n",
-       "                             '2.4e-09', 'num_tokens': '2.522e+06', 'completions/mean_length':                      \n",
-       "                             '41.82', 'completions/min_length': '22', 'completions/max_length':                    \n",
-       "                             '128', 'completions/clipped_ratio': '0.0625',                                         \n",
-       "                             'completions/mean_terminated_length': '36.08',                                        \n",
-       "                             'completions/min_terminated_length': '22',                                            \n",
-       "                             'completions/max_terminated_length': '60',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7812',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.415',                                              \n",
-       "                             'rewards/format_reward/mean': '0.9371',                                               \n",
-       "                             'rewards/format_reward/std': '0.2445', 'reward': '1.25',                              \n",
-       "                             'reward_std': '0.4865', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '3.988e-05', 'entropy': '0.01252', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '16.04', 'epoch': '0.312'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:11:54]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029515;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029516;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m76\u001b[0m%|███████▌  | \u001b[1m38\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:02<02:57, \u001b[1m14.\u001b[0m76s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m78\u001b[0m%|███████▊  |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m39\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:18<02:46, \u001b[1m15.\u001b[0m15s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.002493', 'grad_norm': '2.959', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '2.4e-09', 'num_tokens': '2.522e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '41.82', 'completions/min_length': '22', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '128', 'completions/clipped_ratio': '0.0625',                       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '36.08',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '22',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '60',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7812',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.415',                            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9371',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2445', 'reward': '1.25',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.4865', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '3.988e-05', 'entropy': '0.01252', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '16.04', 'epoch': '0.312'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:12:05] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             78%|███████▊  | 39/50 [09:18<02:46, 15.15s/it]#015 80%|████████  |                    \n",
-       "                             40/50 [09:30<02:23, 14.38s/it]#015                                                    \n",
-       "                             #015{'loss': '0.007042', 'grad_norm': '3.101', 'learning_rate':                       \n",
-       "                             '2.2e-09', 'num_tokens': '2.585e+06', 'completions/mean_length':                      \n",
-       "                             '36.74', 'completions/min_length': '22', 'completions/max_length':                    \n",
-       "                             '59', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '36.74',                                        \n",
-       "                             'completions/min_terminated_length': '22',                                            \n",
-       "                             'completions/max_terminated_length': '59',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.8281',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.3788',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9607',                                               \n",
-       "                             'rewards/format_reward/std': '0.1955', 'reward': '1.308',                             \n",
-       "                             'reward_std': '0.431', 'frac_reward_zero_std': '0', 'kl':                             \n",
-       "                             '0.0004843', 'entropy': '0.1136', 'clip_ratio/low_mean': '0',                         \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '12.57', 'epoch': '0.32'}                                                \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:12:05]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029523;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029524;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m78\u001b[0m%|███████▊  | \u001b[1m39\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:18<02:46, \u001b[1m15.\u001b[0m15s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m80\u001b[0m%|████████  |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m40\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:30<02:23, \u001b[1m14.\u001b[0m38s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.007042', 'grad_norm': '3.101', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '2.2e-09', 'num_tokens': '2.585e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '36.74', 'completions/min_length': '22', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '59', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '36.74',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '22',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '59',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8281',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.3788',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9607',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.1955', 'reward': '1.308',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.431', 'frac_reward_zero_std': '0', 'kl':           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0004843', 'entropy': '0.1136', 'clip_ratio/low_mean': '0',       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '12.57', 'epoch': '0.32'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:12:21] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             80%|████████  | 40/50 [09:30<02:23, 14.38s/it]#015 82%|████████▏ |                    \n",
-       "                             41/50 [09:46<02:12, 14.71s/it]#015                                                    \n",
-       "                             #015{'loss': '0.001328', 'grad_norm': '1.698', 'learning_rate':                       \n",
-       "                             '2e-09', 'num_tokens': '2.646e+06', 'completions/mean_length':                        \n",
-       "                             '31.11', 'completions/min_length': '19', 'completions/max_length':                    \n",
-       "                             '77', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '31.11',                                        \n",
-       "                             'completions/min_terminated_length': '19',                                            \n",
-       "                             'completions/max_terminated_length': '77',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7734',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4203',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9058',                                               \n",
-       "                             'rewards/format_reward/std': '0.2941', 'reward': '1.226',                             \n",
-       "                             'reward_std': '0.5212', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '2.695e-05', 'entropy': '0.01587', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '15.47', 'epoch': '0.328'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:12:21]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029531;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029532;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m80\u001b[0m%|████████  | \u001b[1m40\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:30<02:23, \u001b[1m14.\u001b[0m38s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m82\u001b[0m%|████████▏ |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m41\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:46<02:12, \u001b[1m14.\u001b[0m71s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.001328', 'grad_norm': '1.698', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '2e-09', 'num_tokens': '2.646e+06', 'completions/mean_length':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '31.11', 'completions/min_length': '19', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '77', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '31.11',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '19',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '77',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7734',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4203',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9058',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2941', 'reward': '1.226',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5212', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '2.695e-05', 'entropy': '0.01587', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '15.47', 'epoch': '0.328'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:12:32] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             82%|████████▏ | 41/50 [09:46<02:12, 14.71s/it]#015 84%|████████▍ |                    \n",
-       "                             42/50 [09:58<01:50, 13.84s/it]#015                                                    \n",
-       "                             #015{'loss': '-0.007931', 'grad_norm': '2.842', 'learning_rate':                      \n",
-       "                             '1.8e-09', 'num_tokens': '2.724e+06', 'completions/mean_length':                      \n",
-       "                             '32.8', 'completions/min_length': '22', 'completions/max_length':                     \n",
-       "                             '45', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '32.8',                                         \n",
-       "                             'completions/min_terminated_length': '22',                                            \n",
-       "                             'completions/max_terminated_length': '45',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.8281',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.3788',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9371',                                               \n",
-       "                             'rewards/format_reward/std': '0.2445', 'reward': '1.297',                             \n",
-       "                             'reward_std': '0.4592', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '0.0001043', 'entropy': '0.02917', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '11.8', 'epoch': '0.336'}                                                \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:12:32]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029539;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029540;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m82\u001b[0m%|████████▏ | \u001b[1m41\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:46<02:12, \u001b[1m14.\u001b[0m71s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m84\u001b[0m%|████████▍ |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m42\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:58<01:50, \u001b[1m13.\u001b[0m84s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.007931', 'grad_norm': '2.842', 'learning_rate':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '1.8e-09', 'num_tokens': '2.724e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '32.8', 'completions/min_length': '22', 'completions/max_length':   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '45', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '32.8',                       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '22',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '45',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8281',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.3788',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9371',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2445', 'reward': '1.297',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.4592', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0001043', 'entropy': '0.02917', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '11.8', 'epoch': '0.336'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:12:48] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             84%|████████▍ | 42/50 [09:58<01:50, 13.84s/it]#015 86%|████████▌ |                    \n",
-       "                             43/50 [10:10<01:32, 13.24s/it]#015                                                    \n",
-       "                             #015{'loss': '-0.00114', 'grad_norm': '3.581', 'learning_rate':                       \n",
-       "                             '1.6e-09', 'num_tokens': '2.783e+06', 'completions/mean_length':                      \n",
-       "                             '26.7', 'completions/min_length': '20', 'completions/max_length':                     \n",
-       "                             '48', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '26.7',                                         \n",
-       "                             'completions/min_terminated_length': '20',                                            \n",
-       "                             'completions/max_terminated_length': '48',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7656',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4253',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9293',                                               \n",
-       "                             'rewards/format_reward/std': '0.258', 'reward': '1.23',                               \n",
-       "                             'reward_std': '0.502', 'frac_reward_zero_std': '0', 'kl':                             \n",
-       "                             '0.0001331', 'entropy': '0.01795', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '11.82', 'epoch': '0.344'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:12:48]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029547;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029548;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m84\u001b[0m%|████████▍ | \u001b[1m42\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m09:58<01:50, \u001b[1m13.\u001b[0m84s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m86\u001b[0m%|████████▌ |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m43\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:10<01:32, \u001b[1m13.\u001b[0m24s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.00114', 'grad_norm': '3.581', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '1.6e-09', 'num_tokens': '2.783e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '26.7', 'completions/min_length': '20', 'completions/max_length':   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '48', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '26.7',                       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '48',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7656',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4253',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9293',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.258', 'reward': '1.23',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.502', 'frac_reward_zero_std': '0', 'kl':           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0001331', 'entropy': '0.01795', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '11.82', 'epoch': '0.344'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:12:59] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             86%|████████▌ | 43/50 [10:10<01:32, 13.24s/it]#015 88%|████████▊ |                    \n",
-       "                             44/50 [10:24<01:21, 13.65s/it]#015                                                    \n",
-       "                             #015{'loss': '0.002971', 'grad_norm': '2.827', 'learning_rate':                       \n",
-       "                             '1.4e-09', 'num_tokens': '2.851e+06', 'completions/mean_length':                      \n",
-       "                             '28.79', 'completions/min_length': '19', 'completions/max_length':                    \n",
-       "                             '40', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '28.79',                                        \n",
-       "                             'completions/min_terminated_length': '19',                                            \n",
-       "                             'completions/max_terminated_length': '40',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7266',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4475',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9372',                                               \n",
-       "                             'rewards/format_reward/std': '0.2442', 'reward': '1.195',                             \n",
-       "                             'reward_std': '0.511', 'frac_reward_zero_std': '0', 'kl':                             \n",
-       "                             '0.0001698', 'entropy': '0.03675', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '14.59', 'epoch': '0.352'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:12:59]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029555;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029556;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m86\u001b[0m%|████████▌ | \u001b[1m43\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:10<01:32, \u001b[1m13.\u001b[0m24s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m88\u001b[0m%|████████▊ |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m44\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:24<01:21, \u001b[1m13.\u001b[0m65s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.002971', 'grad_norm': '2.827', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '1.4e-09', 'num_tokens': '2.851e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '28.79', 'completions/min_length': '19', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '40', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '28.79',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '19',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '40',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7266',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4475',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9372',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2442', 'reward': '1.195',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.511', 'frac_reward_zero_std': '0', 'kl':           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0001698', 'entropy': '0.03675', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '14.59', 'epoch': '0.352'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:13:15] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             88%|████████▊ | 44/50 [10:24<01:21, 13.65s/it]#015 90%|█████████ |                    \n",
-       "                             45/50 [10:36<01:05, 13.02s/it]#015                                                    \n",
-       "                             #015{'loss': '0.001591', 'grad_norm': '3.238', 'learning_rate':                       \n",
-       "                             '1.2e-09', 'num_tokens': '2.911e+06', 'completions/mean_length':                      \n",
-       "                             '32.55', 'completions/min_length': '21', 'completions/max_length':                    \n",
-       "                             '45', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '32.55',                                        \n",
-       "                             'completions/min_terminated_length': '21',                                            \n",
-       "                             'completions/max_terminated_length': '45',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.8984',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.3033',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9136',                                               \n",
-       "                             'rewards/format_reward/std': '0.2829', 'reward': '1.355',                             \n",
-       "                             'reward_std': '0.4362', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '5.299e-06', 'entropy': '0.009558', 'clip_ratio/low_mean': '0',                       \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '11.53', 'epoch': '0.36'}                                                \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:13:15]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029563;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029564;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m88\u001b[0m%|████████▊ | \u001b[1m44\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:24<01:21, \u001b[1m13.\u001b[0m65s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m90\u001b[0m%|█████████ |  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m45\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:36<01:05, \u001b[1m13.\u001b[0m02s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.001591', 'grad_norm': '3.238', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '1.2e-09', 'num_tokens': '2.911e+06', 'completions/mean_length':    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '32.55', 'completions/min_length': '21', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '45', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '32.55',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '21',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '45',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.8984',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.3033',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9136',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2829', 'reward': '1.355',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.4362', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '5.299e-06', 'entropy': '0.009558', 'clip_ratio/low_mean': '0',     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '11.53', 'epoch': '0.36'\u001b[1m}\u001b[0m                              \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:13:26] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             90%|█████████ | 45/50 [10:36<01:05, 13.02s/it]#015 92%|█████████▏|                    \n",
-       "                             46/50 [10:48<00:51, 12.95s/it]#015                                                    \n",
-       "                             #015{'loss': '0.009148', 'grad_norm': '2.514', 'learning_rate':                       \n",
-       "                             '1e-09', 'num_tokens': '2.973e+06', 'completions/mean_length':                        \n",
-       "                             '34.28', 'completions/min_length': '21', 'completions/max_length':                    \n",
-       "                             '69', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '34.28',                                        \n",
-       "                             'completions/min_terminated_length': '21',                                            \n",
-       "                             'completions/max_terminated_length': '69',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.6328',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4839',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9529',                                               \n",
-       "                             'rewards/format_reward/std': '0.2133', 'reward': '1.109',                             \n",
-       "                             'reward_std': '0.525', 'frac_reward_zero_std': '0', 'kl':                             \n",
-       "                             '0.0002607', 'entropy': '0.0535', 'clip_ratio/low_mean': '0',                         \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '12.78', 'epoch': '0.368'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:13:26]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029571;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029572;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m90\u001b[0m%|█████████ | \u001b[1m45\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:36<01:05, \u001b[1m13.\u001b[0m02s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m92\u001b[0m%|█████████▏|  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m46\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:48<00:51, \u001b[1m12.\u001b[0m95s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.009148', 'grad_norm': '2.514', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '1e-09', 'num_tokens': '2.973e+06', 'completions/mean_length':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '34.28', 'completions/min_length': '21', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '69', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '34.28',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '21',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '69',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.6328',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4839',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9529',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.2133', 'reward': '1.109',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.525', 'frac_reward_zero_std': '0', 'kl':           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0002607', 'entropy': '0.0535', 'clip_ratio/low_mean': '0',       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '12.78', 'epoch': '0.368'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:13:42] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             92%|█████████▏| 46/50 [10:48<00:51, 12.95s/it]#015 94%|█████████▍|                    \n",
-       "                             47/50 [11:03<00:40, 13.34s/it]#015                                                    \n",
-       "                             #015{'loss': '0.003021', 'grad_norm': '2.457', 'learning_rate':                       \n",
-       "                             '8e-10', 'num_tokens': '3.033e+06', 'completions/mean_length':                        \n",
-       "                             '33.32', 'completions/min_length': '20', 'completions/max_length':                    \n",
-       "                             '55', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '33.32',                                        \n",
-       "                             'completions/min_terminated_length': '20',                                            \n",
-       "                             'completions/max_terminated_length': '55',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.7266',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4475',                                             \n",
-       "                             'rewards/format_reward/mean': '0.8743',                                               \n",
-       "                             'rewards/format_reward/std': '0.334', 'reward': '1.164',                              \n",
-       "                             'reward_std': '0.5659', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '1.138e-05', 'entropy': '0.01608', 'clip_ratio/low_mean': '0',                        \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '14.23', 'epoch': '0.376'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:13:42]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029579;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029580;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m92\u001b[0m%|█████████▏| \u001b[1m46\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m10:48<00:51, \u001b[1m12.\u001b[0m95s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m94\u001b[0m%|█████████▍|  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m47\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:03<00:40, \u001b[1m13.\u001b[0m34s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.003021', 'grad_norm': '2.457', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '8e-10', 'num_tokens': '3.033e+06', 'completions/mean_length':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '33.32', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '55', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '33.32',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '55',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7266',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4475',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8743',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.334', 'reward': '1.164',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5659', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '1.138e-05', 'entropy': '0.01608', 'clip_ratio/low_mean': '0',      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '14.23', 'epoch': '0.376'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:13:58] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             94%|█████████▍| 47/50 [11:03<00:40, 13.34s/it]#015 96%|█████████▌|                    \n",
-       "                             48/50 [11:19<00:28, 14.08s/it]#015                                                    \n",
-       "                             #015{'loss': '0.007249', 'grad_norm': '1.409', 'learning_rate':                       \n",
-       "                             '6e-10', 'num_tokens': '3.098e+06', 'completions/mean_length':                        \n",
-       "                             '37.45', 'completions/min_length': '20', 'completions/max_length':                    \n",
-       "                             '128', 'completions/clipped_ratio': '0.0625',                                         \n",
-       "                             'completions/mean_terminated_length': '31.41',                                        \n",
-       "                             'completions/min_terminated_length': '20',                                            \n",
-       "                             'completions/max_terminated_length': '100',                                           \n",
-       "                             'rewards/tool_call_reward/mean': '0.7422',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4391',                                             \n",
-       "                             'rewards/format_reward/mean': '0.8739',                                               \n",
-       "                             'rewards/format_reward/std': '0.3349', 'reward': '1.179',                             \n",
-       "                             'reward_std': '0.5614', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '4.697e-05', 'entropy': '0.0159', 'clip_ratio/low_mean': '0',                         \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '15.81', 'epoch': '0.384'}                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:13:58]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029587;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029588;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m94\u001b[0m%|█████████▍| \u001b[1m47\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:03<00:40, \u001b[1m13.\u001b[0m34s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m96\u001b[0m%|█████████▌|  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m48\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:19<00:28, \u001b[1m14.\u001b[0m08s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.007249', 'grad_norm': '1.409', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '6e-10', 'num_tokens': '3.098e+06', 'completions/mean_length':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '37.45', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '128', 'completions/clipped_ratio': '0.0625',                       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '31.41',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '100',                         \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7422',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4391',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.8739',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.3349', 'reward': '1.179',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.5614', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '4.697e-05', 'entropy': '0.0159', 'clip_ratio/low_mean': '0',       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '15.81', 'epoch': '0.384'\u001b[1m}\u001b[0m                             \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:14:09] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             96%|█████████▌| 48/50 [11:19<00:28, 14.08s/it]#015 98%|█████████▊|                    \n",
-       "                             49/50 [11:35<00:14, 14.71s/it]#015                                                    \n",
-       "                             #015{'loss': '-0.02007', 'grad_norm': '2.001', 'learning_rate':                       \n",
-       "                             '4e-10', 'num_tokens': '3.167e+06', 'completions/mean_length':                        \n",
-       "                             '37.2', 'completions/min_length': '21', 'completions/max_length':                     \n",
-       "                             '109', 'completions/clipped_ratio': '0',                                              \n",
-       "                             'completions/mean_terminated_length': '37.2',                                         \n",
-       "                             'completions/min_terminated_length': '21',                                            \n",
-       "                             'completions/max_terminated_length': '109',                                           \n",
-       "                             'rewards/tool_call_reward/mean': '0.7344',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.4434',                                             \n",
-       "                             'rewards/format_reward/mean': '0.882', 'rewards/format_reward/std':                   \n",
-       "                             '0.3253', 'reward': '1.175', 'reward_std': '0.5572',                                  \n",
-       "                             'frac_reward_zero_std': '0', 'kl': '8.196e-05', 'entropy':                            \n",
-       "                             '0.0125', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',                      \n",
-       "                             'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',                              \n",
-       "                             'clip_ratio/region_mean': '0', 'step_time': '16.16', 'epoch':                         \n",
-       "                             '0.392'}                                                                              \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:14:09]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029595;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029596;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m96\u001b[0m%|█████████▌| \u001b[1m48\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:19<00:28, \u001b[1m14.\u001b[0m08s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m \u001b[1m98\u001b[0m%|█████████▊|  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:35<00:14, \u001b[1m14.\u001b[0m71s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '-0.02007', 'grad_norm': '2.001', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '4e-10', 'num_tokens': '3.167e+06', 'completions/mean_length':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '37.2', 'completions/min_length': '21', 'completions/max_length':   \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '109', 'completions/clipped_ratio': '0',                            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '37.2',                       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '21',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '109',                         \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.7344',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.4434',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.882', 'rewards/format_reward/std': \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.3253', 'reward': '1.175', 'reward_std': '0.5572',                \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'frac_reward_zero_std': '0', 'kl': '8.196e-05', 'entropy':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.0125', 'clip_ratio/low_mean': '0', 'clip_ratio/low_min': '0',    \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_mean': '0', 'clip_ratio/high_max': '0',            \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/region_mean': '0', 'step_time': '16.16', 'epoch':       \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '0.392'\u001b[1m}\u001b[0m                                                            \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:14:25] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             98%|█████████▊| 49/50 [11:35<00:14, 14.71s/it]#015100%|██████████|                    \n",
-       "                             50/50 [11:50<00:00, 14.97s/it]#015                                                    \n",
-       "                             #015{'loss': '0.004886', 'grad_norm': '3.012', 'learning_rate':                       \n",
-       "                             '2e-10', 'num_tokens': '3.236e+06', 'completions/mean_length':                        \n",
-       "                             '29.63', 'completions/min_length': '20', 'completions/max_length':                    \n",
-       "                             '46', 'completions/clipped_ratio': '0',                                               \n",
-       "                             'completions/mean_terminated_length': '29.63',                                        \n",
-       "                             'completions/min_terminated_length': '20',                                            \n",
-       "                             'completions/max_terminated_length': '46',                                            \n",
-       "                             'rewards/tool_call_reward/mean': '0.9219',                                            \n",
-       "                             'rewards/tool_call_reward/std': '0.2694',                                             \n",
-       "                             'rewards/format_reward/mean': '0.9686',                                               \n",
-       "                             'rewards/format_reward/std': '0.1756', 'reward': '1.406',                             \n",
-       "                             'reward_std': '0.3309', 'frac_reward_zero_std': '0', 'kl':                            \n",
-       "                             '6.488e-05', 'entropy': '0.009783', 'clip_ratio/low_mean': '0',                       \n",
-       "                             'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',                               \n",
-       "                             'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',                            \n",
-       "                             'step_time': '15.57', 'epoch': '0.4'}                                                 \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:14:25]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029603;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029604;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m98\u001b[0m%|█████████▊| \u001b[1m49\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:35<00:14, \u001b[1m14.\u001b[0m71s/it\u001b[1m]\u001b[0m#\u001b[1m015100\u001b[0m%|██████████|  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m50\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:50<00:00, \u001b[1m14.\u001b[0m97s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                                  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'loss': '0.004886', 'grad_norm': '3.012', 'learning_rate':     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '2e-10', 'num_tokens': '3.236e+06', 'completions/mean_length':      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '29.63', 'completions/min_length': '20', 'completions/max_length':  \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '46', 'completions/clipped_ratio': '0',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/mean_terminated_length': '29.63',                      \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/min_terminated_length': '20',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'completions/max_terminated_length': '46',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/mean': '0.9219',                          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/tool_call_reward/std': '0.2694',                           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/mean': '0.9686',                             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'rewards/format_reward/std': '0.1756', 'reward': '1.406',           \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'reward_std': '0.3309', 'frac_reward_zero_std': '0', 'kl':          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         '6.488e-05', 'entropy': '0.009783', 'clip_ratio/low_mean': '0',     \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/low_min': '0', 'clip_ratio/high_mean': '0',             \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'clip_ratio/high_max': '0', 'clip_ratio/region_mean': '0',          \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'step_time': '15.57', 'epoch': '0.4'\u001b[1m}\u001b[0m                               \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             100%|██████████| 50/50 [11:50<00:00, 14.97s/it]#015                                   \n",
-       "                             #015{'train_runtime': '710.8', 'train_samples_per_second': '9.004',                   \n",
-       "                             'train_steps_per_second': '0.07', 'train_loss': '0.003163',                           \n",
-       "                             'epoch': '0.4'}                                                                       \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029611;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029612;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m100\u001b[0m%|██████████| \u001b[1m50\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:50<00:00, \u001b[1m14.\u001b[0m97s/it\u001b[1m]\u001b[0m#\u001b[1m015\u001b[0m                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         #\u001b[1m015\u001b[0m\u001b[1m{\u001b[0m'train_runtime': '710.8', 'train_samples_per_second': '9.004', \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'train_steps_per_second': '0.07', 'train_loss': '0.003163',         \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         'epoch': '0.4'\u001b[1m}\u001b[0m                                                     \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             100%|██████████| 50/50 [11:50<00:00, 14.97s/it]#015100%|██████████|                   \n",
-       "                             50/50 [11:50<00:00, 14.22s/it]                                                        \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029619;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029620;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m100\u001b[0m%|██████████| \u001b[1m50\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:50<00:00, \u001b[1m14.\u001b[0m97s/it\u001b[1m]\u001b[0m#\u001b[1m015100\u001b[0m%|██████████| \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m50\u001b[0m/\u001b[1m50\u001b[0m \u001b[1m[\u001b[0m11:50<00:00, \u001b[1m14.\u001b[0m22s/it\u001b[1m]\u001b[0m                                      \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:14:36] INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             Writing model shards:   0%|          | 0/1 [00:00<?,                                  \n",
-       "                             ?it/s]#015Writing model shards: 100%|██████████| 1/1 [00:02<00:00,                    \n",
-       "                             2.73s/it]#015Writing model shards: 100%|██████████| 1/1                               \n",
-       "                             [00:02<00:00,  2.73s/it]                                                              \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:14:36]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029627;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029628;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Writing model shards:   \u001b[1m0\u001b[0m%|          | \u001b[1m0\u001b[0m/\u001b[1m1\u001b[0m \u001b[1m[\u001b[0m00:00                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             ++ echo 'Training Container Execution Completed'                                      \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029635;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029636;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         ++ echo 'Training Container Execution Completed'                    \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     smolm3-grpo-toolcall-20260626-085316-20260626105316/algo-1-17824642 resources.py:31439\n",
-       "                             49:                                                                                   \n",
-       "                             Training Container Execution Completed                                                \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m smolm3-grpo-toolcall-\u001b[1m20260626\u001b[0m-\u001b[1m085316\u001b[0m-\u001b[1m20260626105316\u001b[0m/algo-\u001b[1m1\u001b[0m-\u001b[1m17824642\u001b[0m \u001b]8;id=17029643;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029644;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31439\u001b\\\u001b[2m31439\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[1m49\u001b[0m:                                                                 \u001b[2m                  \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         Training Container Execution Completed                              \u001b[2m                  \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/26/26 11:14:57] INFO     Final Resource Status: Completed                                    resources.py:31442\n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/26/26 11:14:57]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m Final Resource Status: \u001b[1mCompleted\u001b[0m                                    \u001b]8;id=17029651;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py\u001b\\\u001b[2mresources.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=17029652;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/sagemaker/core/resources.py#31442\u001b\\\u001b[2m31442\u001b[0m\u001b]8;;\u001b\\\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
\n"
-      ],
-      "text/plain": []
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    }
-   ],
+   "outputs": [],
    "source": [
     "import time\n",
     "from sagemaker.train.model_trainer import ModelTrainer\n",
@@ -6600,7 +1075,7 @@
    "source": [
     "import boto3\n",
     "\n",
-    "TRAINING_JOB_NAME = \"\"  # set to a specific completed job name, or leave blank for latest\n",
+    "TRAINING_JOB_NAME = \"smolm3-grpo-toolcall-20260626-085316-20260626105316\"  # set to a specific completed job name, or leave blank for latest\n",
     "JOB_PREFIX = \"smolm3-grpo-toolcall-\"\n",
     "\n",
     "sm = boto3.client(\"sagemaker\", region_name=REGION)\n",
@@ -6792,6 +1267,16 @@
     "plt.show()\n"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "id": "c1648ee9",
+   "metadata": {},
+   "source": [
+    "![GRPO reward curves](training_reward_curves.png)\n",
+    "\n",
+    "Read the curve as a setup check. The exact-match and format rewards should stay away from zero; a sudden flatline near zero, especially with many clipped completions, usually means generation collapsed.\n"
+   ]
+  },
   {
    "cell_type": "markdown",
    "id": "5338ef32",
@@ -6803,6 +1288,22 @@
     "\n",
     "The run above is deliberately short so the notebook stays practical to execute. To turn it into a real training run, train for more steps, use a held-out evaluation split, and track exact-match tool-call accuracy in addition to the reward curves.\n"
    ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "1e9f6063",
+   "metadata": {},
+   "source": [
+    "## References\n",
+    "\n",
+    "- [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](https://arxiv.org/abs/2501.12948)\n",
+    "- [Hugging Face Hub authentication](https://huggingface.co/docs/huggingface_hub/v1.21.0.rc0/en/package_reference/authentication#huggingface_hub.notebook_login)\n",
+    "- [TRL GRPO Trainer](https://huggingface.co/docs/trl/main/en/grpo_trainer)\n",
+    "- [SageMaker `ModelTrainer`](https://sagemaker.readthedocs.io/en/stable/api/training/model_trainer.html)\n",
+    "- [DeepSpeed ZeRO](https://www.deepspeed.ai/tutorials/zero/)\n",
+    "- [`HuggingFaceTB/SmolLM3-3B`](https://huggingface.co/HuggingFaceTB/SmolLM3-3B)\n",
+    "- [`Salesforce/xlam-function-calling-60k`](https://huggingface.co/datasets/Salesforce/xlam-function-calling-60k)\n"
+   ]
   }
  ],
  "metadata": {

From f80c46e076153168cac260c6b4d9b3c153ae66be Mon Sep 17 00:00:00 2001
From: DWarez 
Date: Wed, 1 Jul 2026 17:49:34 +0200
Subject: [PATCH 5/5] fix: now relying on public image

Signed-off-by: DWarez 
---
 .../grpo-llm-trl/sagemaker-notebook.ipynb     | 437 ++++--------------
 1 file changed, 94 insertions(+), 343 deletions(-)

diff --git a/docs/sagemaker/notebooks/sagemaker-sdk/grpo-llm-trl/sagemaker-notebook.ipynb b/docs/sagemaker/notebooks/sagemaker-sdk/grpo-llm-trl/sagemaker-notebook.ipynb
index d0b73408a..c2906332e 100644
--- a/docs/sagemaker/notebooks/sagemaker-sdk/grpo-llm-trl/sagemaker-notebook.ipynb
+++ b/docs/sagemaker/notebooks/sagemaker-sdk/grpo-llm-trl/sagemaker-notebook.ipynb
@@ -84,26 +84,10 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 1,
+   "execution_count": null,
    "id": "e1a0499a",
    "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "HF token loaded\n"
-     ]
-    },
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "/Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
-      "  from .autonotebook import tqdm as notebook_tqdm\n"
-     ]
-    }
-   ],
+   "outputs": [],
    "source": [
     "from huggingface_hub import get_token, notebook_login\n",
     "\n",
@@ -117,41 +101,10 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 2,
+   "execution_count": null,
    "id": "954730c2",
    "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "sagemaker.config INFO - Not applying SDK defaults from location: /Library/Application Support/sagemaker/config.yaml\n",
-      "sagemaker.config INFO - Not applying SDK defaults from location: /Users/dwarez/Library/Application Support/sagemaker/config.yaml\n"
-     ]
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/27/26 12:50:38] INFO     Loading cached SSO token for hf                                          tokens.py:312\n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/27/26 12:50:38]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m Loading cached SSO token for hf                                          \u001b]8;id=8266086;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/botocore/tokens.py\u001b\\\u001b[2mtokens.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=8266087;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/botocore/tokens.py#312\u001b\\\u001b[2m312\u001b[0m\u001b]8;;\u001b\\\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "region: us-east-1\n",
-      "bucket: sagemaker-us-east-1-754289655784\n",
-      "role:   arn:aws:iam::754289655784:role/aws-reserved/sso.amazonaws.com/AWSReservedSSO_HF-Sandbox-access_a9a3037b77bf6782\n"
-     ]
-    }
-   ],
+   "outputs": [],
    "source": [
     "import boto3\n",
     "from sagemaker.core.helper.session_helper import Session, get_execution_role\n",
@@ -177,32 +130,92 @@
    "source": [
     "## Configuration\n",
     "\n",
-    "`MODEL_ID` is the base model. `INSTANCE_TYPE` controls the training hardware. `TRAINING_IMAGE` is the ECR image SageMaker runs.\n",
-    "\n"
+    "`MODEL_ID` is the base model and `INSTANCE_TYPE` the training hardware. `PUBLIC_IMAGE` is the public image that carries the TRL/GRPO training stack; the next step mirrors it into a private ECR repo (`ECR_REPO`) that SageMaker can actually pull from, and sets `TRAINING_IMAGE` to that private URI."
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 3,
-   "id": "f54550f3-220b-438b-aeba-d593bdb44538",
+   "execution_count": null,
+   "id": "7280ac2e",
    "metadata": {},
    "outputs": [],
    "source": [
-    "role = \"arn:aws:iam::754289655784:role/sagemaker_execution_role\""
+    "MODEL_ID = \"HuggingFaceTB/SmolLM3-3B\"      # SmolLM3, HF's 3B instruct model\n",
+    "INSTANCE_TYPE = \"ml.p4d.24xlarge\"          # 8 x A100 40GB\n",
+    "\n",
+    "# Public image carrying the TRL/GRPO training stack. SageMaker can't pull a\n",
+    "# public registry directly (see the next cell), so we mirror it into a private\n",
+    "# ECR repo in your account and train from that.\n",
+    "PUBLIC_IMAGE = \"public.ecr.aws/u9a4y4p1/huggingface/hubcap:aws-pytorch-training_2.11.0-transformers5.10.2-gpu-py312-cu130-amzn2023-sagemaker\"\n",
+    "ECR_REPO = \"hf-trl-grpo\"                    # private repo created in your account"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "9a688d8f",
+   "metadata": {},
+   "source": [
+    "## Copy the training image to a private ECR repo\n",
+    "\n",
+    "SageMaker only pulls from private ECR, so the cell below mirrors the public image into your account once and sets `TRAINING_IMAGE`. Expand it if you want the details."
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 4,
-   "id": "7280ac2e",
+   "execution_count": null,
+   "id": "8e777f4e",
    "metadata": {},
    "outputs": [],
    "source": [
-    "MODEL_ID = \"HuggingFaceTB/SmolLM3-3B\"      # SmolLM3, HF's 3B instruct model\n",
-    "INSTANCE_TYPE = \"ml.p4d.24xlarge\"          # 8 x A100 40GB\n",
+    "import base64\n",
+    "import platform\n",
+    "import shutil\n",
+    "import subprocess\n",
+    "import tarfile\n",
+    "import tempfile\n",
+    "import urllib.request\n",
+    "from pathlib import Path\n",
     "\n",
-    "# TODO: Temporary review image; replace with the public tutorial image before publishing.\n",
-    "TRAINING_IMAGE = \"754289655784.dkr.ecr.us-east-1.amazonaws.com/hf-trl-grpo:sagemaker-trl-dev-e63f67e\""
+    "account_id = boto3.client(\"sts\", region_name=REGION).get_caller_identity()[\"Account\"]\n",
+    "registry = f\"{account_id}.dkr.ecr.{REGION}.amazonaws.com\"\n",
+    "tag = PUBLIC_IMAGE.rsplit(\":\", 1)[-1]\n",
+    "TRAINING_IMAGE = f\"{registry}/{ECR_REPO}:{tag}\"\n",
+    "\n",
+    "\n",
+    "def ensure_crane():\n",
+    "    \"\"\"Path to a `crane` binary, downloading a release build if it isn't on PATH.\"\"\"\n",
+    "    if found := shutil.which(\"crane\"):\n",
+    "        return found\n",
+    "    system = platform.system()                                     # Darwin / Linux\n",
+    "    machine = {\"aarch64\": \"arm64\"}.get(platform.machine(), platform.machine())  # x86_64 / arm64\n",
+    "    asset = f\"go-containerregistry_{system}_{machine}.tar.gz\"\n",
+    "    url = f\"https://github.com/google/go-containerregistry/releases/latest/download/{asset}\"\n",
+    "    dest = Path(tempfile.gettempdir()) / \"crane-bin\"\n",
+    "    dest.mkdir(exist_ok=True)\n",
+    "    urllib.request.urlretrieve(url, dest / asset)\n",
+    "    with tarfile.open(dest / asset) as tf:\n",
+    "        tf.extract(\"crane\", dest)\n",
+    "    crane = dest / \"crane\"\n",
+    "    crane.chmod(0o755)\n",
+    "    return str(crane)\n",
+    "\n",
+    "\n",
+    "# 1. Create the private repo (no-op if it already exists).\n",
+    "ecr = boto3.client(\"ecr\", region_name=REGION)\n",
+    "try:\n",
+    "    ecr.create_repository(repositoryName=ECR_REPO)\n",
+    "except ecr.exceptions.RepositoryAlreadyExistsException:\n",
+    "    pass\n",
+    "\n",
+    "# 2. Authenticate crane to your private ECR (the public source needs no login).\n",
+    "token = ecr.get_authorization_token()[\"authorizationData\"][0][\"authorizationToken\"]\n",
+    "password = base64.b64decode(token).decode().split(\":\", 1)[1]\n",
+    "crane = ensure_crane()\n",
+    "subprocess.run([crane, \"auth\", \"login\", registry, \"-u\", \"AWS\", \"-p\", password], check=True)\n",
+    "\n",
+    "# 3. Copy public -> private. Idempotent: only missing layers are transferred.\n",
+    "subprocess.run([crane, \"copy\", PUBLIC_IMAGE, TRAINING_IMAGE], check=True)\n",
+    "print(\"training image:\", TRAINING_IMAGE)"
    ]
   },
   {
@@ -221,184 +234,10 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 5,
+   "execution_count": null,
    "id": "2690d04e",
    "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "text/html": [
-       "
[06/27/26 12:51:04] INFO     HTTP Request: GET https://huggingface.co/api/agent-harnesses \"HTTP/1.1 _client.py:1025\n",
-       "                             200 OK\"                                                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/27/26 12:51:04]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m HTTP Request: \u001b[1mGET\u001b[0m \u001b[4mhttps://huggingface.co/api/agent-harnesses\u001b[0m \"HTTP/1.1 \u001b]8;id=8266096;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py\u001b\\\u001b[2m_client.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=8266097;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py#1025\u001b\\\u001b[2m1025\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         200 OK\"                                                                \u001b[2m               \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     HTTP Request: HEAD                                                     _client.py:1025\n",
-       "                             https://huggingface.co/datasets/Salesforce/xlam-function-calling-60k/r                \n",
-       "                             esolve/main/README.md \"HTTP/1.1 200 OK\"                                               \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m HTTP Request: \u001b[1mHEAD\u001b[0m                                                     \u001b]8;id=8266104;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py\u001b\\\u001b[2m_client.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=8266105;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py#1025\u001b\\\u001b[2m1025\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[4mhttps://huggingface.co/datasets/Salesforce/xlam-function-calling-60k/r\u001b[0m \u001b[2m               \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[4mesolve/main/README.md\u001b[0m \"HTTP/1.1 200 OK\"                                \u001b[2m               \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     HTTP Request: HEAD                                                     _client.py:1025\n",
-       "                             https://huggingface.co/datasets/Salesforce/xlam-function-calling-60k/r                \n",
-       "                             esolve/26d14ebfe18b1f7b524bd39b404b50af5dc97866/xlam-function-calling-                \n",
-       "                             60k.py \"HTTP/1.1 404 Not Found\"                                                       \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m HTTP Request: \u001b[1mHEAD\u001b[0m                                                     \u001b]8;id=8266112;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py\u001b\\\u001b[2m_client.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=8266113;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py#1025\u001b\\\u001b[2m1025\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[4mhttps://huggingface.co/datasets/Salesforce/xlam-function-calling-60k/r\u001b[0m \u001b[2m               \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[4mesolve/26d14ebfe18b1f7b524bd39b404b50af5dc97866/xlam-function-calling-\u001b[0m \u001b[2m               \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[4m60k.py\u001b[0m \"HTTP/1.1 404 Not Found\"                                        \u001b[2m               \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     HTTP Request: HEAD                                                     _client.py:1025\n",
-       "                             https://s3.amazonaws.com/datasets.huggingface.co/datasets/datasets/Sal                \n",
-       "                             esforce/xlam-function-calling-60k/Salesforce/xlam-function-calling-60k                \n",
-       "                             .py \"HTTP/1.1 404 Not Found\"                                                          \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m HTTP Request: \u001b[1mHEAD\u001b[0m                                                     \u001b]8;id=8266120;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py\u001b\\\u001b[2m_client.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=8266121;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py#1025\u001b\\\u001b[2m1025\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[4mhttps://s3.amazonaws.com/datasets.huggingface.co/datasets/datasets/Sal\u001b[0m \u001b[2m               \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[4mesforce/xlam-function-calling-60k/Salesforce/xlam-function-calling-60k\u001b[0m \u001b[2m               \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[4m.py\u001b[0m \"HTTP/1.1 404 Not Found\"                                           \u001b[2m               \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     HTTP Request: GET                                                      _client.py:1025\n",
-       "                             https://huggingface.co/api/datasets/Salesforce/xlam-function-calling-6                \n",
-       "                             0k/revision/26d14ebfe18b1f7b524bd39b404b50af5dc97866 \"HTTP/1.1 200 OK\"                \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m HTTP Request: \u001b[1mGET\u001b[0m                                                      \u001b]8;id=8266128;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py\u001b\\\u001b[2m_client.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=8266129;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py#1025\u001b\\\u001b[2m1025\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[4mhttps://huggingface.co/api/datasets/Salesforce/xlam-function-calling-6\u001b[0m \u001b[2m               \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[4m0k/revision/26d14ebfe18b1f7b524bd39b404b50af5dc97866\u001b[0m \"HTTP/1.1 200 OK\" \u001b[2m               \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
[06/27/26 12:51:05] INFO     HTTP Request: HEAD                                                     _client.py:1025\n",
-       "                             https://huggingface.co/datasets/Salesforce/xlam-function-calling-60k/r                \n",
-       "                             esolve/26d14ebfe18b1f7b524bd39b404b50af5dc97866/.huggingface.yaml                     \n",
-       "                             \"HTTP/1.1 404 Not Found\"                                                              \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m[06/27/26 12:51:05]\u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m HTTP Request: \u001b[1mHEAD\u001b[0m                                                     \u001b]8;id=8266136;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py\u001b\\\u001b[2m_client.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=8266137;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py#1025\u001b\\\u001b[2m1025\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[4mhttps://huggingface.co/datasets/Salesforce/xlam-function-calling-60k/r\u001b[0m \u001b[2m               \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[4mesolve/26d14ebfe18b1f7b524bd39b404b50af5dc97866/.huggingface.yaml\u001b[0m      \u001b[2m               \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"HTTP/1.1 404 Not Found\"                                               \u001b[2m               \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     HTTP Request: GET                                                      _client.py:1025\n",
-       "                             https://datasets-server.huggingface.co/info?dataset=Salesforce/xlam-fu                \n",
-       "                             nction-calling-60k \"HTTP/1.1 200 OK\"                                                  \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m HTTP Request: \u001b[1mGET\u001b[0m                                                      \u001b]8;id=8266144;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py\u001b\\\u001b[2m_client.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=8266145;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py#1025\u001b\\\u001b[2m1025\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[4mhttps://datasets-server.huggingface.co/info?\u001b[0m\u001b[4mdataset\u001b[0m\u001b[4m=\u001b[0m\u001b[4mSalesforce\u001b[0m\u001b[4m/xlam-fu\u001b[0m \u001b[2m               \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[4mnction-calling-60k\u001b[0m \"HTTP/1.1 200 OK\"                                   \u001b[2m               \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     HTTP Request: GET                                                      _client.py:1025\n",
-       "                             https://huggingface.co/api/datasets/Salesforce/xlam-function-calling-6                \n",
-       "                             0k/tree/26d14ebfe18b1f7b524bd39b404b50af5dc97866?recursive=false&expan                \n",
-       "                             d=false \"HTTP/1.1 200 OK\"                                                             \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m HTTP Request: \u001b[1mGET\u001b[0m                                                      \u001b]8;id=8266152;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py\u001b\\\u001b[2m_client.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=8266153;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py#1025\u001b\\\u001b[2m1025\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[4mhttps://huggingface.co/api/datasets/Salesforce/xlam-function-calling-6\u001b[0m \u001b[2m               \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[4m0k/tree/26d14ebfe18b1f7b524bd39b404b50af5dc97866?\u001b[0m\u001b[4mrecursive\u001b[0m\u001b[4m=\u001b[0m\u001b[4mfalse\u001b[0m\u001b[4m&\u001b[0m\u001b[4mexpan\u001b[0m \u001b[2m               \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[4md\u001b[0m\u001b[4m=\u001b[0m\u001b[4mfalse\u001b[0m \"HTTP/1.1 200 OK\"                                              \u001b[2m               \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/html": [
-       "
                    INFO     HTTP Request: HEAD                                                     _client.py:1025\n",
-       "                             https://huggingface.co/datasets/Salesforce/xlam-function-calling-60k/r                \n",
-       "                             esolve/26d14ebfe18b1f7b524bd39b404b50af5dc97866/dataset_infos.json                    \n",
-       "                             \"HTTP/1.1 404 Not Found\"                                                              \n",
-       "
\n"
-      ],
-      "text/plain": [
-       "\u001b[2m                   \u001b[0m\u001b[2m \u001b[0m\u001b[1mINFO    \u001b[0m HTTP Request: \u001b[1mHEAD\u001b[0m                                                     \u001b]8;id=8266160;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py\u001b\\\u001b[2m_client.py\u001b[0m\u001b]8;;\u001b\\\u001b[2m:\u001b[0m\u001b]8;id=8266161;file:///Users/dwarez/hf/projects/agents-trainin-docs/.venv/lib/python3.14/site-packages/httpx/_client.py#1025\u001b\\\u001b[2m1025\u001b[0m\u001b]8;;\u001b\\\n",
-       "\u001b[2m                    \u001b[0m         \u001b[4mhttps://huggingface.co/datasets/Salesforce/xlam-function-calling-60k/r\u001b[0m \u001b[2m               \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \u001b[4mesolve/26d14ebfe18b1f7b524bd39b404b50af5dc97866/dataset_infos.json\u001b[0m     \u001b[2m               \u001b[0m\n",
-       "\u001b[2m                    \u001b[0m         \"HTTP/1.1 404 Not Found\"                                               \u001b[2m               \u001b[0m\n"
-      ]
-     },
-     "metadata": {},
-     "output_type": "display_data"
-    },
-    {
-     "data": {
-      "text/plain": [
-       "{'query': 'Where can I find live giveaways for beta access and games?',\n",
-       " 'tools': '[{\"name\": \"live_giveaways_by_type\", \"description\": \"Retrieve live giveaways from the GamerPower API based on the specified type.\", \"parameters\": {\"type\": {\"description\": \"The type of giveaways to retrieve (e.g., game, loot, beta).\", \"type\": \"str\", \"default\": \"game\"}}}]',\n",
-       " 'answers': '[{\"name\": \"live_giveaways_by_type\", \"arguments\": {\"type\": \"beta\"}}, {\"name\": \"live_giveaways_by_type\", \"arguments\": {\"type\": \"game\"}}]'}"
-      ]
-     },
-     "execution_count": 5,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
+   "outputs": [],
    "source": [
     "from datasets import load_dataset\n",
     "\n",
@@ -418,31 +257,10 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 7,
+   "execution_count": null,
    "id": "2b17c98c",
    "metadata": {},
-   "outputs": [
-    {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "Saving the dataset (1/1 shards): 100%|██████████| 2000/2000 [00:00<00:00, 421665.23 examples/s]\n"
-     ]
-    },
-    {
-     "data": {
-      "text/plain": [
-       "Dataset({\n",
-       "    features: ['answers', 'prompt'],\n",
-       "    num_rows: 2000\n",
-       "})"
-      ]
-     },
-     "execution_count": 7,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
+   "outputs": [],
    "source": [
     "import shutil\n",
     "from pathlib import Path\n",
@@ -495,23 +313,10 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 8,
+   "execution_count": null,
    "id": "c15e88de",
    "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "[\n",
-      "  {\n",
-      "    \"role\": \"system\",\n",
-      "    \"content\": \"/no_think\\nYou are an expert in composing function calls. Return exactly one function call that answers the user's request.\\n\\nYou have access to the following tools:\\n\\n[{\\\"name\\\": \\\"user_metadata_information\\\", \\\"description\\\": \\\"Fetch and return metadata information for a specified TikTok user, such as number of followers, followings, avatar URL, description, and more.\\\", \\\"parameters\\\": {\\\"username\\\": {\\\"description\\\": \\\"The TikTok username to fetch metadata for (e.g., \\\\\\\"amazon\\\\\\\").\\\", \\\"type\\\": \\\"str\\\", \\\"default\\\": \\\"tiktok\\\"}, \\\"fresh\\\": {\\\"description\\\": \\\"If set to '1', forces the API to return fresh (non-cached) data. Defaults to '0'.\\\", \\\"type\\\": \\\"str, optional\\\", \\\"default\\\": \\\"0\\\"}}}, {\\\"name\\\": \\\"medias\\\", \\\"description\\\": \\\"Retrieves media posts from a specified Instagram user using the Instagram RapidAPI.\\\", \\\"parameters\\\": {\\\"user_id\\\": {\\\"description\\\": \\\"The ID of the Instagram user whose media posts are to be retrieved.\\\", \\\"type\\\": \\\"str\\\", \\\"default\\\": \\\"25025320\\\"}, \\\"batch_size\\\": {\\\"description\\\": \\\"The number of media posts to retrieve in a single batch, ranging from 1 to 50. Defaults to 20.\\\", \\\"type\\\": \\\"int, optional\\\", \\\"default\\\": \\\"20\\\"}, \\\"next_cursor\\\": {\\\"description\\\": \\\"The cursor for fetching the next set of media posts in pagination. Defaults to None.\\\", \\\"type\\\": \\\"str, optional\\\", \\\"default\\\": \\\"\\\"}}}, {\\\"name\\\": \\\"followers\\\", \\\"description\\\": \\\"Retrieves the list of followers for\n",
-      "\n",
-      "ground truth: [{\"name\": \"followers\", \"arguments\": {\"user_id\": \"17841420039002923\", \"batch_size\": 25}}]\n"
-     ]
-    }
-   ],
+   "outputs": [],
    "source": [
     "print(json.dumps(prepared[0][\"prompt\"], indent=2)[:1500])\n",
     "print(\"\\nground truth:\", prepared[0][\"answers\"])"
@@ -527,18 +332,10 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 9,
+   "execution_count": null,
    "id": "0859c513",
    "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "s3://sagemaker-us-east-1-754289655784/xlam-grpo/prepared\n"
-     ]
-    }
-   ],
+   "outputs": [],
    "source": [
     "train_s3 = sess.upload_data(\"prepared\", bucket=bucket, key_prefix=\"xlam-grpo/prepared\")\n",
     "print(train_s3)"
@@ -561,7 +358,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 10,
+   "execution_count": null,
    "id": "8dab8b5e",
    "metadata": {},
    "outputs": [],
@@ -572,18 +369,10 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 11,
+   "execution_count": null,
    "id": "11c358f0",
    "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Overwriting src/rewards.py\n"
-     ]
-    }
-   ],
+   "outputs": [],
    "source": [
     "%%writefile src/rewards.py\n",
     "'''Verifiable rewards for GRPO tool-call training.\n",
@@ -721,21 +510,10 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 12,
+   "execution_count": null,
    "id": "d91edf68",
    "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "parsed:      [{'name': 'get_weather', 'arguments': {'city': 'Paris', 'unit': 'c'}}]\n",
-      "exact-match: [1.0]\n",
-      "format:      [1.0]\n",
-      "wrong call:  [0.0]\n"
-     ]
-    }
-   ],
+   "outputs": [],
    "source": [
     "from src.rewards import parse_tool_calls, tool_call_reward, format_reward\n",
     "\n",
@@ -767,18 +545,10 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 13,
+   "execution_count": null,
    "id": "f42a218b",
    "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Overwriting src/train.py\n"
-     ]
-    }
-   ],
+   "outputs": [],
    "source": [
     "%%writefile src/train.py\n",
     "'''SageMaker entry script: GRPO tool-call training with TRL.\n",
@@ -950,18 +720,10 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 14,
+   "execution_count": null,
    "id": "395b0169",
    "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Overwriting src/ds_zero3.json\n"
-     ]
-    }
-   ],
+   "outputs": [],
    "source": [
     "%%writefile src/ds_zero3.json\n",
     "{\n",
@@ -1075,7 +837,7 @@
    "source": [
     "import boto3\n",
     "\n",
-    "TRAINING_JOB_NAME = \"smolm3-grpo-toolcall-20260626-085316-20260626105316\"  # set to a specific completed job name, or leave blank for latest\n",
+    "TRAINING_JOB_NAME = \"\"  # set to a specific completed job name, or leave blank for latest\n",
     "JOB_PREFIX = \"smolm3-grpo-toolcall-\"\n",
     "\n",
     "sm = boto3.client(\"sagemaker\", region_name=REGION)\n",
@@ -1267,16 +1029,6 @@
     "plt.show()\n"
    ]
   },
-  {
-   "cell_type": "markdown",
-   "id": "c1648ee9",
-   "metadata": {},
-   "source": [
-    "![GRPO reward curves](training_reward_curves.png)\n",
-    "\n",
-    "Read the curve as a setup check. The exact-match and format rewards should stay away from zero; a sudden flatline near zero, especially with many clipped completions, usually means generation collapsed.\n"
-   ]
-  },
   {
    "cell_type": "markdown",
    "id": "5338ef32",
@@ -1297,7 +1049,6 @@
     "## References\n",
     "\n",
     "- [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](https://arxiv.org/abs/2501.12948)\n",
-    "- [Hugging Face Hub authentication](https://huggingface.co/docs/huggingface_hub/v1.21.0.rc0/en/package_reference/authentication#huggingface_hub.notebook_login)\n",
     "- [TRL GRPO Trainer](https://huggingface.co/docs/trl/main/en/grpo_trainer)\n",
     "- [SageMaker `ModelTrainer`](https://sagemaker.readthedocs.io/en/stable/api/training/model_trainer.html)\n",
     "- [DeepSpeed ZeRO](https://www.deepspeed.ai/tutorials/zero/)\n",
@@ -1322,7 +1073,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.14.5"
+   "version": "3.14.6"
   }
  },
  "nbformat": 4,