Apple most likely is never fixing the fp16 attention bug. (comfyanonymous#8485)

Author: comfyanonymous

comfy/model_management.py

@@ -1052,7 +1052,7 @@ def pytorch_attention_flash_attention():
     global ENABLE_PYTORCH_ATTENTION
     if ENABLE_PYTORCH_ATTENTION:
         #TODO: more reliable way of checking for flash attention?
-        if is_nvidia(): #pytorch flash attention only works on Nvidia
+        if is_nvidia():
             return True
         if is_intel_xpu():
             return True
@@ -1068,7 +1068,7 @@ def force_upcast_attention_dtype():
     upcast = args.force_upcast_attention
 
     macos_version = mac_version()
-    if macos_version is not None and ((14, 5) <= macos_version < (16,)):  # black image bug on recent versions of macOS
+    if macos_version is not None and ((14, 5) <= macos_version):  # black image bug on recent versions of macOS, I don't think it's ever getting fixed
         upcast = True
 
     if upcast: