youngkingdom/vllm
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

AITER MHA off by default

Browse Source 
Signed-off-by: Micah Williamson <micah.williamson@amd.com>
This commit is contained in:
Micah Williamson
2025-10-20 15:48:36 +00:00
committed by Lu Fang
parent 2072fdc044
commit c7021f1270
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/envs.py
View File

@ -105,7 +105,7 @@ if TYPE_CHECKING:
VLLM_ROCM_USE_AITER_MOE: bool = True
VLLM_ROCM_USE_AITER_RMSNORM: bool = True
VLLM_ROCM_USE_AITER_MLA: bool = True
VLLM_ROCM_USE_AITER_MHA: bool = True
VLLM_ROCM_USE_AITER_MHA: bool = False
VLLM_ROCM_USE_AITER_FP4_ASM_GEMM: bool = False
VLLM_ROCM_USE_TRITON_ROPE: bool = False
VLLM_ROCM_USE_AITER_FP8BMM: bool = True
@ -862,7 +862,7 @@ environment_variables: dict[str, Callable[[], Any]] = {
# Whether to use aiter mha ops.
# By default is enabled.
"VLLM_ROCM_USE_AITER_MHA": lambda: (
os.getenv("VLLM_ROCM_USE_AITER_MHA", "True").lower() in ("true", "1")
os.getenv("VLLM_ROCM_USE_AITER_MHA", "False").lower() in ("true", "1")
),
# Whether to use aiter fp4 gemm asm.
# By default is disabled.