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.

Use basic attention implementation for small inputs on old pytorch.

comfyanonymous6a7bc35db845
diff --git a/comfy/ldm/modules/attention.py b/comfy/ldm/modules/attention.py
index 309240d5c..fd8888d0e 100644
--- a/comfy/ldm/modules/attention.py
+++ b/comfy/ldm/modules/attention.py
@@ -351,8 +351,11 @@ else:
optimized_attention_masked = optimized_attention
def optimized_attention_for_device(device, mask=False, small_input=False):
- if small_input and model_management.pytorch_attention_enabled():
- return attention_pytorch #TODO: need to confirm but this is probably slightly faster for small inputs in all cases
+ if small_input:
+ if model_management.pytorch_attention_enabled():
+ return attention_pytorch #TODO: need to confirm but this is probably slightly faster for small inputs in all cases
+ else:
+ return attention_basic
if device == torch.device("cpu"):
return attention_sub_quad