xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-09 04:45:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Rocm] Set VLLM_ROCM_USE_AITER_FUSION_SHARED_EXPERTS default is disabled (#28985)

Browse Source 
Signed-off-by: zhyajie <yajizhan@amd.com>
Co-authored-by: zhyajie <yajizhan@amd.com>
Co-authored-by: TJian <tunjian.tan@embeddedllm.com>
This commit is contained in:
杰兮 2025-11-28 18:08:42 +08:00 committed by GitHub
parent ccbdf51bd5
commit 3cb32e5d6e
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 3 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
vllm/envs.py
View File

@ -116,7 +116,7 @@ if TYPE_CHECKING:
VLLM_ROCM_USE_AITER_TRITON_ROPE: bool = False VLLM_ROCM_USE_AITER_TRITON_ROPE: bool = False
VLLM_ROCM_USE_AITER_FP8BMM: bool = True VLLM_ROCM_USE_AITER_FP8BMM: bool = True
VLLM_ROCM_USE_AITER_UNIFIED_ATTENTION: bool = False VLLM_ROCM_USE_AITER_UNIFIED_ATTENTION: bool = False
VLLM_ROCM_USE_AITER_FUSION_SHARED_EXPERTS: bool = True VLLM_ROCM_USE_AITER_FUSION_SHARED_EXPERTS: bool = False
VLLM_ROCM_USE_AITER_TRITON_GEMM: bool = True VLLM_ROCM_USE_AITER_TRITON_GEMM: bool = True
VLLM_ROCM_USE_SKINNY_GEMM: bool = True VLLM_ROCM_USE_SKINNY_GEMM: bool = True
VLLM_ROCM_FP8_PADDING: bool = True VLLM_ROCM_FP8_PADDING: bool = True
@ -969,9 +969,9 @@ environment_variables: dict[str, Callable[[], Any]] = {
in ("true", "1") in ("true", "1")
), ),
# Whether to use aiter fusion shared experts ops. # Whether to use aiter fusion shared experts ops.
# By default is enabled. # By default is disabled.
"VLLM_ROCM_USE_AITER_FUSION_SHARED_EXPERTS": lambda: ( "VLLM_ROCM_USE_AITER_FUSION_SHARED_EXPERTS": lambda: (
os.getenv("VLLM_ROCM_USE_AITER_FUSION_SHARED_EXPERTS", "True").lower() os.getenv("VLLM_ROCM_USE_AITER_FUSION_SHARED_EXPERTS", "False").lower()
in ("true", "1") in ("true", "1")
), ),
# Whether to use aiter triton kernels for gemm ops. # Whether to use aiter triton kernels for gemm ops.