Skip to content

The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.

Enable pytorch attention by default on xpu.

comfyanonymous321c5fa2958a
diff --git a/comfy/model_management.py b/comfy/model_management.py
index e38ef4eea..d8bc3bfea 100644
--- a/comfy/model_management.py
+++ b/comfy/model_management.py
@@ -165,6 +165,9 @@ try:
ENABLE_PYTORCH_ATTENTION = True
if torch.cuda.is_bf16_supported():
VAE_DTYPE = torch.bfloat16
+ if is_intel_xpu():
+ if args.use_split_cross_attention == False and args.use_quad_cross_attention == False:
+ ENABLE_PYTORCH_ATTENTION = True
except:
pass