kanghui0204

[Bug]: some questions regarding the usage of NCCL allreduce/broadcast/allgather/send/recv in VLLM using pycomm and torch's distributed.

1

### Your current environment The output of `python collect_env.py` ```text Your output of `python collect_env.py` here ``` ### 🐛 Describe the bug Hello VLLM expert, While reading the VLLM code,...

feat: Low Precision Allreduce for PCIe based GPU

21

last PR：https://github.com/NVIDIA/TensorRT-LLM/pull/3851 last revet PR：https://github.com/NVIDIA/TensorRT-LLM/pull/4340

**Description** I've identified a bug in the latest version of FBGEMM where the set_optimizer_step() function no longer resets the Adam optimizer iteration counter, breaking previous functionality that worked in earlier...