feat(experimental): add bitwise Qwen3-MoE train–inference parity case based on GRPO + vLLM(TP>1) + FSDP - #430

feat(experimental): add bitwise Qwen3-MoE train–inference parity case based on GRPO + vLLM(TP>1) + FSDP#430

cboss6 wants to merge 2 commits into

Tencent-Hunyuan:mainTencent-Hunyuan/UniRL:mainfrom

cboss6:qwen3-moe-parity-tp4cboss6/UniRL:qwen3-moe-parity-tp4Copy head branch name to clipboard