Repository Issue Activity (beta)

#499 [Bug] 权重同步未应用模型自带的 hf_to_vllm_mapper，wrapper 形态模型（如 model.language_model.* 前缀）在第一次权重更新即失败

#474 如何使用cot格式对Qwen3.5-0.8B进行全量微调

#496 求，什么时候支持一下RSI。

#495 BatchStratifiedSampler crashes with ZeroDivisionError when a domain_ratios entry has no matching rows

#423 Bug: image_grid_thw[image_index][0] IndexError: list index out of range

#490 [RFC] Upstream Environment-Regularized Policy Optimization (ERPO) into ROLL

#485 Colocated full-parameter weight sync: CPU fallback costs ~17% of every training step on seccomp-managed containers; request a pidfd_getfd-free GPU transport

#484 Megatron colocated LoRA weight sync fails on CUDA IPC and needs CPU staging fallback

#482 Verify evals on Papers with Code

#475 [Bug] Advantage whitening fails with a single valid response token

#329 Training hangs at actor-infer step with Qwen3-8B on an 8-GPU node

#316 AttributeError: 'NoneType' object has no attribute 'rename_key_'

#411 Error: Qwen3.5-35B-A3B lora sft with mcore-adapter, error occurs when saving a ckpt.

#419 mcore-adapter从VLM的hfconfig中获取参数异常

#150 deepspeed zero3 model_update error

#418 ValueError: There is no module or parameter named 'base_model' in Qwen2_5_VLForConditionalGeneration

#442 LR scheduler progress can be inconsistent with dynamic batching in Megatron actor training

#435 [Bug] Training/Inference crash with Qwen2/3-VL due to missing mm_token_type_ids in Collator

#205 间断显卡编号的设备映射问题（English translation: device_mapping cannot handle discontinuous GPU indexes）

#279 weight update progress过于缓慢

#309 code_dapo agentic task fails

#394 Very slow asynchronous GRPO training on 8×A100

#230 ROLL 是否可以加速reward模型（非LLM）计算与LLM GRPO的高效协同训练

#407 LR scheduler exhausts early in agentic training with AgentNativeStepEnvManager

#472 [RFC] Add MindSpeed/Megatron Unit Test CI Pipeline