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 13, 17:00 UTC
[CI/Build] Update checking logic in cutlass_group_gemm_supported (#27948)
Signed-off-by: zhewenli <zhewenli@meta.com>
Zhewen Li committed
Nov 5, 2025 at 15:40 UTC
5ee93a5956e82a260b9944d2f0cf42f0f225204a
1 file changed
+5
-1
vllm/_custom_ops.py
+5
-1
@@ -836,7 +836,11 @@ def cutlass_sparse_scaled_mm_supported(cuda_device_capability: int) -> bool:
836
837
838
def cutlass_group_gemm_supported(cuda_device_capability: int) -> bool:
839
- return torch.ops._C.cutlass_group_gemm_supported(cuda_device_capability)
839
+ try:
840
+ return torch.ops._C.cutlass_group_gemm_supported(cuda_device_capability)
841
+ except AttributeError:
842
+ # Return False on non-CUDA platforms where it is not available
843
+ return False
844
845
846
def cutlass_sparse_compress(a: torch.Tensor) -> tuple[torch.Tensor, torch.Tensor]: