fix local windowed attention to be overlapping

Author: lucidrains

alphafold3_pytorch/attention.py

@@ -374,11 +374,11 @@ def local_attn(
         # just do radius of 1 for now
         # perhaps not even necessary, and could try shifted windows (a la Swin)
 
-        k, v = tuple(pad_at_dim(t, (1, 0), dim = -2) for t in (k, v))
-        mask = F.pad(mask, (1, 0), value = False)
+        k, v = tuple(pad_at_dim(t, (1, 0), dim = -3) for t in (k, v))
+        mask = pad_at_dim(mask, (1, 0), dim = -2, value = False)
 
-        k, v = tuple(torch.cat((t[..., :-1, :], t[..., 1:, :]), dim = -2) for t in (k, v))
-        mask = torch.cat((mask[..., :-1], mask[..., 1:]), dim = -1)
+        k, v = tuple(torch.cat((t[..., :-1, :, :], t[..., 1:, :, :]), dim = -2) for t in (k, v))
+        mask = torch.cat((mask[..., :-1, :], mask[..., 1:, :]), dim = -1)
 
         # handle attention bias (inefficiently)
 

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "alphafold3-pytorch"
-version = "0.6.6"
+version = "0.6.7"
 description = "Alphafold 3 - Pytorch"
 authors = [
     { name = "Phil Wang", email = "[email protected]" },