feat: dynamo inference backend integration#2737

Open

biswapanda wants to merge 19 commits into

PrimeIntellect-ai:mainfrom

biswapanda:rl-sdk-4

Commits on Jun 9, 2026

feat(dynamo): AdminAPI abstraction + dynamo backend selector + RL worker discovery
biswapanda
committed
feat(weight-transfer): NCCL broadcast + FP8/E8M0 conversion for GB200 (qwen3_moe/glm_moe)
biswapanda
committed
feat(broadcast): NFS-safe filesystem broadcast + weight_broadcast.keep_recent for LoRA
biswapanda
committed
feat(orchestrator): dispatch compute_teacher_logprobs by renderer_transport (vllm + dynamo nvext)
biswapanda
committed
feat(inference): vLLM 0.22 patches (fp32 lm-head, int64 silu_mul_quant, padded scrub)
biswapanda
committed
build(deps): point verifiers/renderers submodules at biswapanda rl-sdk-4 forks (TITO)
biswapanda
committed
build(image): Dockerfile.cuda.runtime (vLLM 0.22, COPY deps, DeepGEMM) + Dockerfile.dynamo
biswapanda
committed
feat(deploy): helm chart updates + dynamo k8s manifests + smoke-test tools
biswapanda
committed
fix(rl): rename renderer_transport values to vllm_generate/dynamo_chat; bump verifiers/renderers deps to rl-sdk-4 heads
biswapanda
committed