diff --git a/comfy/model_management.py b/comfy/model_management.py
index 017df17d1..7dcfca35d 100644
--- a/comfy/model_management.py
+++ b/comfy/model_management.py
@@ -1966,7 +1966,7 @@ def supports_nvfp4_compute(device=None):
     # cuBLAS FP4 matmul kernels require CUDA 13+, see #11864
     try:
         cuda_version_major = int(torch.version.cuda.split(".")[0])
-    except:
+    except (AttributeError, ValueError):
         return False
     if cuda_version_major < 13:
         return False