x-transformers 1.42.15__py3-none-any.whl → 1.42.17__py3-none-any.whl

x_transformers/x_transformers.py

@@ -1235,9 +1235,9 @@ class Attention(Module):
         # maybe learned value residual mixer per token
 
         self.to_value_residual_mix = nn.Sequential(
-            nn.Linear(dim, 1),
+            nn.Linear(dim, heads),
             nn.Sigmoid(),
-            Rearrange('b n 1 -> b 1 n 1')
+            Rearrange('b n h -> b h n 1')
          ) if learned_value_residual_mix else always(0.5)
 
         # attention on attention
@@ -1428,13 +1428,15 @@ class Attention(Module):
             else:
                 attn_bias = rel_pos(i, j)
 
-            attn_bias = pad_at_dim(attn_bias, (num_mem_kv, 0), value = 0.) # handle memory key / values
+            attn_bias = pad_at_dim(attn_bias, (num_mem_kv, 0)) # handle memory key / values
 
         # prepare data dependent alibi from forgetting transformers paper, if needed
 
         if exists(self.data_dependent_alibi):
             attn_bias = self.data_dependent_alibi(x)
 
+            attn_bias = pad_at_dim(attn_bias, (num_mem_kv, 0))
+
         # attention is all we need
 
         out, intermediates = self.attend(

{x_transformers-1.42.15.dist-info → x_transformers-1.42.17.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: x-transformers
-Version: 1.42.15
+Version: 1.42.17
 Summary: X-Transformers - Pytorch
 Home-page: https://github.com/lucidrains/x-transformers
 Author: Phil Wang

{x_transformers-1.42.15.dist-info → x_transformers-1.42.17.dist-info}/RECORD

@@ -6,11 +6,11 @@ x_transformers/dpo.py,sha256=xt4OuOWhU8pN3OKN2LZAaC2NC8iiEnchqqcrPWVqf0o,3521
 x_transformers/multi_input.py,sha256=tCh-fTJDj2ib4SMGtsa-AM8MxKzJAQSwqAXOu3HU2mg,9252
 x_transformers/neo_mlp.py,sha256=XCNnnop9WLarcxap1kGuYc1x8GHvwkZiDRnXOxSl3Po,3452
 x_transformers/nonautoregressive_wrapper.py,sha256=2NU58hYMgn-4Jzg3mie-mXb0XH_dCN7fjlzd3K1rLUY,10510
-x_transformers/x_transformers.py,sha256=-gi7UiCRdp-5y34cUJEMk7uFSi-I7khXxON1gErAKbY,95125
+x_transformers/x_transformers.py,sha256=Wvkw4j_78413LdCnCt_wHgcVFiCbzrC8u4TH2iXhkNU,95181
 x_transformers/xl_autoregressive_wrapper.py,sha256=CvZMJ6A6PA-Y_bQAhnORwjJBSl6Vjq2IdW5KTdk8NI8,4195
 x_transformers/xval.py,sha256=7S00kCuab4tWQa-vf-z-XfzADjVj48MoFIr7VSIvttg,8575
-x_transformers-1.42.15.dist-info/LICENSE,sha256=As9u198X-U-vph5noInuUfqsAG2zX_oXPHDmdjwlPPY,1066
-x_transformers-1.42.15.dist-info/METADATA,sha256=zqzIQ3mdFjs4WV7IgTu4YYEmFM-6GKWast8twY4__Tg,717
-x_transformers-1.42.15.dist-info/WHEEL,sha256=PZUExdf71Ui_so67QXpySuHtCi3-J3wvF4ORK6k_S8U,91
-x_transformers-1.42.15.dist-info/top_level.txt,sha256=hO6KGpFuGucRNEtRfme4A_rGcM53AKwGP7RVlRIxS5Q,15
-x_transformers-1.42.15.dist-info/RECORD,,
+x_transformers-1.42.17.dist-info/LICENSE,sha256=As9u198X-U-vph5noInuUfqsAG2zX_oXPHDmdjwlPPY,1066
+x_transformers-1.42.17.dist-info/METADATA,sha256=T1MDXNdxqdPkqFpGuJVb7vBhniGCbHefm5C-lhb3LJk,717
+x_transformers-1.42.17.dist-info/WHEEL,sha256=PZUExdf71Ui_so67QXpySuHtCi3-J3wvF4ORK6k_S8U,91
+x_transformers-1.42.17.dist-info/top_level.txt,sha256=hO6KGpFuGucRNEtRfme4A_rGcM53AKwGP7RVlRIxS5Q,15
+x_transformers-1.42.17.dist-info/RECORD,,