Update attention.py

ldm/modules/attention.py

@@ -174,23 +174,30 @@ class CrossAttention(nn.Module):
         context = default(context, x)
         k = self.to_k(context)
         v = self.to_v(context)
+        del context, x
 
         q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> (b h) n d', h=h), (q, k, v))
 
-        sim = einsum('b i d, b j d -> b i j', q, k) * self.scale
+        r1 = torch.zeros(q.shape[0], q.shape[1], v.shape[2], device=q.device)
 
-        if exists(mask):
-            mask = rearrange(mask, 'b ... -> b (...)')
-            max_neg_value = -torch.finfo(sim.dtype).max
-            mask = repeat(mask, 'b j -> (b h) () j', h=h)
-            sim.masked_fill_(~mask, max_neg_value)
+        # valid values for steps = 2,4,8,16,32,64
+        # higher steps is slower but less memory usage
+        # at 16 can run 1920x1536 on a 3090, at 64 can run over 1920x1920
+        # speed seems to be impacted more on 30x series cards
+        steps = 16
+        slice_size = q.shape[1] // steps if q.shape[1] % steps == 0 else q.shape[1]
+        for i in range(0, q.shape[1], slice_size):
+            end = i + slice_size
+            s1 = einsum('b i d, b j d -> b i j', q[:, i:end], k)
+            s1 *= self.scale
+            s2 = s1.softmax(dim=-1)
+            del s1
+            r1[:, i:end] = einsum('b i j, b j d -> b i d', s2, v)
+            del s2
+        r2 = rearrange(r1, '(b h) n d -> b n (h d)', h=h)
+        del r1
 
-        # attention, what we cannot get enough of
-        attn = sim.softmax(dim=-1)
-
-        out = einsum('b i j, b j d -> b i d', attn, v)
-        out = rearrange(out, '(b h) n d -> b n (h d)', h=h)
-        return self.to_out(out)
+        return self.to_out(r2)
 
 
 class BasicTransformerBlock(nn.Module):