This makes pytorch2.0 attention perform a bit faster.

comfy/ldm/modules/attention.py

@@ -455,11 +455,7 @@ class CrossAttentionPytorch(nn.Module):
 
         b, _, _ = q.shape
         q, 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 NotImplementedError
         out = (
-            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)