--disable-xformers should not even try to import xformers.

comfy/ldm/modules/attention.py

@@ -11,11 +11,10 @@ from .sub_quadratic_attention import efficient_dot_product_attention
 
 import model_management
 
-try:
+
+if model_management.xformers_enabled():
     import xformers
     import xformers.ops
-except:
-    pass
 
 # CrossAttn precision handling
 import os

comfy/ldm/modules/diffusionmodules/model.py

@@ -9,11 +9,9 @@ from typing import Optional, Any
 from ldm.modules.attention import MemoryEfficientCrossAttention
 import model_management
 
-try:
+if model_management.xformers_enabled():
     import xformers
     import xformers.ops
-except:
-    pass
 
 try:
     OOM_EXCEPTION = torch.cuda.OutOfMemoryError

comfy/model_management.py

@@ -31,15 +31,16 @@ try:
 except:
     pass
 
-try:
-    import xformers
-    import xformers.ops
-    XFORMERS_IS_AVAILBLE = True
-except:
-    XFORMERS_IS_AVAILBLE = False
-
 if "--disable-xformers" in sys.argv:
     XFORMERS_IS_AVAILBLE = False
+else:
+    try:
+        import xformers
+        import xformers.ops
+        XFORMERS_IS_AVAILBLE = True
+    except:
+        XFORMERS_IS_AVAILBLE = False
+
 
 if "--cpu" in sys.argv:
     vram_state = CPU