[Bugfix] prioritize architecture capability before DeepGEMM availability check (#58073)
Signed-off-by: Tony Lin <tony.lin@intel.com>
Tony Lin committed
Sep 22, 2026 at 23:38 UTC
c42f5285ec9851dacfa370e163e7837e87fd7b6d
1 file changed
+1
-1
vllm/utils/deep_gemm.py
+1
-1
@@ -112,7 +112,7 @@ def is_deep_gemm_supported() -> bool:
112
Currently, only Hopper and Blackwell GPUs are supported.
113
"""
114
is_supported_arch = current_platform.support_deep_gemm()
115
- return envs.VLLM_USE_DEEP_GEMM and has_deep_gemm() and is_supported_arch
115
+ return is_supported_arch and envs.VLLM_USE_DEEP_GEMM and has_deep_gemm()
116
117
118
@functools.cache