[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>

vllm/envs.py

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