perf(kimi-k3): make the KDA temporal state dtype configurable

b82994e

Select commit

Loading

Failed to load commit list.

Uh oh!

There was an error while loading. Please reload this page.

Pre Checkin on: pull_request

Check Code Style with Black

Check Code Style with Ruff

Validate accuracy catalogs against schema

Run non-GPU unit tests

ATOM SGLang Test on: pull_request

Check Heavy CI Gate

Check Pre Checkin Signal

Download aiter wheel

Accuracy (${{ matrix.model_name }})

ATOM Test on: pull_request 3

Check Heavy CI Gate

Offline inference smoke test

Check Pre Checkin Signal

Load test model configs

Download aiter wheel

Accuracy

Update accuracy dashboard

ATOM vLLM Test on: pull_request

Check Heavy CI Gate

Check Pre Checkin Signal

Download aiter wheel

Accuracy (${{ matrix.display_name }})

ATOM Test on: pull_request_review 14

Check Heavy CI Gate

Offline inference smoke test

Check Pre Checkin Signal

Load test model configs

Download aiter wheel

Accuracy (DeepSeek-V4-Pro, deepseek-ai/DeepSeek-V4-Pro, --kv_cache_dtype fp8 --index_cache_dtype ...

Accuracy (DeepSeek-V4-Pro DEP RCCL, deepseek-ai/DeepSeek-V4-Pro, --kv_cache_dtype fp8 --index_cac...

Accuracy (DeepSeek-V4-Pro MTP, deepseek-ai/DeepSeek-V4-Pro, --kv_cache_dtype fp8 --index_cache_dt...

Accuracy (DeepSeek-V4-Pro DSpark, deepseek-ai/DeepSeek-V4-Pro-DSpark, --kv_cache_dtype fp8 --inde...

Accuracy (gpt-oss-120b, openai/gpt-oss-120b, --kv_cache_dtype fp8 --gpu-memory-utilization 0.5, l...

Accuracy (DeepSeek-R1-0528-FP4 MTP, amd/DeepSeek-R1-0528-MXFP4-MTP-MoEFP4, --kv_cache_dtype fp8 -...

Accuracy (Qwen3-Next-80B-A3B-Thinking, Qwen/Qwen3-Next-80B-A3B-Thinking, -tp 8 --kv_cache_dtype f...

Accuracy (Kimi-K2.7-Code-MXFP4, amd/Kimi-K2.7-Code-MXFP4, --kv_cache_dtype fp8 -tp 4 --trust-remo...

Accuracy (Kimi-K3, moonshotai/Kimi-K3, --kv_cache_dtype fp8 -tp 8 --trust-remote-code --gpu-memor...

Accuracy (Kimi-K3 DSpark, moonshotai/Kimi-K3, --kv_cache_dtype fp8 -tp 8 --trust-remote-code --gp...

Accuracy (GLM-5.2-MXFP4 MTP, amd/GLM-5.2-MXFP4, -tp 4 --online_quant_config '{"global_quant_confi...

Accuracy (Qwen3.5-397B-A17B-MXFP4, amd/Qwen3.5-397B-A17B-MXFP4, --kv_cache_dtype fp8 -tp 4 --no-e...

Accuracy (MiniMax-M3-MXFP4, amd/MiniMax-M3-MXFP4, --kv_cache_dtype fp8 -tp 8 --trust-remote-code ...

Accuracy (GLM-5.3-Flash-kpool-16shot, zai-org/GLM-5.3-Flash, --kv_cache_dtype bf16 -tp 8 --max-mo...

Update accuracy dashboard

ATOM SGLang Test on: pull_request_review 2

Check Heavy CI Gate

Check Pre Checkin Signal

Download aiter wheel

Accuracy (${{ matrix.model_name }})

ATOM vLLM Test on: pull_request_review 2

Check Heavy CI Gate

Check Pre Checkin Signal

Download aiter wheel

Accuracy (${{ matrix.display_name }})

ATOM Test on: pull_request

Check Heavy CI Gate

Offline inference smoke test

Check Pre Checkin Signal

Load test model configs

Download aiter wheel

Accuracy

Update accuracy dashboard

ATOM SGLang Test on: pull_request

Check Heavy CI Gate

Check Pre Checkin Signal

Download aiter wheel

Accuracy (${{ matrix.model_name }})

ATOM vLLM Test on: pull_request

Check Heavy CI Gate

Check Pre Checkin Signal

Download aiter wheel

Accuracy (${{ matrix.display_name }})

perf(kimi-k3): make the KDA temporal state dtype configurable

Uh oh!

Select a check to view from the sidebar