A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 12, 09:45 UTC
[Rocm] Set VLLM_ROCM_USE_AITER_FUSION_SHARED_EXPERTS default is disabled (#28985)
Signed-off-by: zhyajie <yajizhan@amd.com> Co-authored-by: zhyajie <yajizhan@amd.com> Co-authored-by: TJian <tunjian.tan@embeddedllm.com>
杰兮 committed
Nov 28, 2025 at 18:08 UTC
3cb32e5d6e227fe91d93902e302ef6f667308e04
1 file changed
+3
-3
vllm/envs.py
+3
-3
@@ -116,7 +116,7 @@ if TYPE_CHECKING:
116
VLLM_ROCM_USE_AITER_TRITON_ROPE: bool = False
117
VLLM_ROCM_USE_AITER_FP8BMM: bool = True
118
VLLM_ROCM_USE_AITER_UNIFIED_ATTENTION: bool = False
119
- VLLM_ROCM_USE_AITER_FUSION_SHARED_EXPERTS: bool = True
119
+ VLLM_ROCM_USE_AITER_FUSION_SHARED_EXPERTS: bool = False
120
VLLM_ROCM_USE_AITER_TRITON_GEMM: bool = True
121
VLLM_ROCM_USE_SKINNY_GEMM: bool = True
122
VLLM_ROCM_FP8_PADDING: bool = True
@@ -969,9 +969,9 @@ environment_variables: dict[str, Callable[[], Any]] = {
969
in ("true", "1")
970
),
971
# Whether to use aiter fusion shared experts ops.
972
- # By default is enabled.
972
+ # By default is disabled.
973
"VLLM_ROCM_USE_AITER_FUSION_SHARED_EXPERTS": lambda: (
974
- os.getenv("VLLM_ROCM_USE_AITER_FUSION_SHARED_EXPERTS", "True").lower()
974
+ os.getenv("VLLM_ROCM_USE_AITER_FUSION_SHARED_EXPERTS", "False").lower()
975
in ("true", "1")
976
),
977
# Whether to use aiter triton kernels for gemm ops.