From c7021f1270b89bb2e68a6ba7a73818358222d6a6 Mon Sep 17 00:00:00 2001 From: Micah Williamson Date: Mon, 20 Oct 2025 15:48:36 +0000 Subject: [PATCH] AITER MHA off by default Signed-off-by: Micah Williamson --- vllm/envs.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/vllm/envs.py b/vllm/envs.py index a948bf9ae7..d0d2b5c2da 100755 --- a/vllm/envs.py +++ b/vllm/envs.py @@ -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.