The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.
This makes pytorch2.0 attention perform a bit faster.
comfyanonymous6908f9c94992
diff --git a/comfy/ldm/modules/attention.py b/comfy/ldm/modules/attention.pyindex 98dbda635..c27d032a3 100644--- a/comfy/ldm/modules/attention.py+++ b/comfy/ldm/modules/attention.py@@ -455,11 +455,7 @@ class CrossAttentionPytorch(nn.Module):b, _, _ = q.shapeq, k, v = map(- lambda t: t.unsqueeze(3)- .reshape(b, t.shape[1], self.heads, self.dim_head)- .permute(0, 2, 1, 3)- .reshape(b * self.heads, t.shape[1], self.dim_head)- .contiguous(),+ lambda t: t.view(b, -1, self.heads, self.dim_head).transpose(1, 2),(q, k, v),)@@ -468,10 +464,7 @@ class CrossAttentionPytorch(nn.Module):if exists(mask):raise NotImplementedErrorout = (- out.unsqueeze(0)- .reshape(b, self.heads, out.shape[1], self.dim_head)- .permute(0, 2, 1, 3)- .reshape(b, out.shape[1], self.heads * self.dim_head)+ out.transpose(1, 2).reshape(b, -1, self.heads * self.dim_head))return self.to_out(out)