[CI/Build] Update checking logic in cutlass_group_gemm_supported (#27948)

Signed-off-by: zhewenli <zhewenli@meta.com>

vllm/_custom_ops.py

@@ -836,7 +836,11 @@ def cutlass_sparse_scaled_mm_supported(cuda_device_capability: int) -> bool:
 
 
 def cutlass_group_gemm_supported(cuda_device_capability: int) -> bool:
+    try:
         return torch.ops._C.cutlass_group_gemm_supported(cuda_device_capability)
+    except AttributeError:
+        # Return False on non-CUDA platforms where it is not available
+        return False
 
 
 def cutlass_sparse_compress(a: torch.Tensor) -> tuple[torch.Tensor, torch.Tensor]: