[gpt-oss] Add gpt-oss mxfp4 support

This commit is contained in:
2025-08-25 15:31:09 +08:00
parent db7f48eeac
commit a7a0adc854
32 changed files with 4835 additions and 1189 deletions

View File

@@ -2,3 +2,6 @@
1. 支持 `gpt-oss-BF16`:将 `vllm` 目录覆盖到镜像中的 `/opt/conda/lib/python3.10/site-packages/vllm`
2.`code_generator.py` 覆盖到镜像中的 `/opt/conda/lib/python3.10/site-packages/triton/compiler/code_generator.py`
3. 启动时指定`VLLM_ATTENTION_BACKEND=TRITON_ATTN_VLLM_V1`
*此版本改动较大,可能因为接口改动,存在部分模型运行出错的问题。*