x-transformers 1.16.14__py3-none-any.whl → 1.16.15__py3-none-any.whl

x_transformers/attend.py

@@ -138,7 +138,7 @@ class Attend(nn.Module):
 
             if causal:
                 causal_mask = torch.ones((q_len, k_len), dtype = torch.bool, device = device).triu(k_len - q_len + 1)
-                mask = mask | causal_mask
+                mask = mask & ~causal_mask
                 causal = False
 
         # handle alibi positional bias
@@ -153,7 +153,7 @@ class Attend(nn.Module):
             mask_value = -torch.finfo(q.dtype).max
 
             if exists(mask):
-                attn_bias = attn_bias.masked_fill(mask, mask_value // 2)
+                attn_bias = attn_bias.masked_fill(~mask, mask_value // 2)
             elif causal:
                 causal_mask = torch.ones((q_len, k_len), dtype = torch.bool, device = device).triu(k_len - q_len + 1)
                 attn_bias = attn_bias.masked_fill(causal_mask, mask_value // 2)
@@ -163,8 +163,6 @@ class Attend(nn.Module):
             # make it an additive bias here
 
             mask = attn_bias
-        else:
-            mask = ~mask
 
         # Check if there is a compatible device for flash attention
 
@@ -226,7 +224,7 @@ class Attend(nn.Module):
         mask_value = -torch.finfo(dots.dtype).max
 
         if exists(mask):
-            dots = dots.masked_fill(mask, mask_value)
+            dots = dots.masked_fill(~mask, mask_value)
 
         if self.causal:
             i, j = dots.shape[-2:]

x_transformers/x_transformers.py

@@ -815,7 +815,7 @@ class Attention(nn.Module):
             masks.append(sparse_topk_mask)
 
         if len(masks) > 0:
-            final_attn_mask = or_reduce(masks)
+            final_attn_mask = ~or_reduce(masks)
 
         # prepare relative positional bias, if needed
 

{x_transformers-1.16.14.dist-info → x_transformers-1.16.15.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: x-transformers
-Version: 1.16.14
+Version: 1.16.15
 Summary: X-Transformers - Pytorch
 Home-page: https://github.com/lucidrains/x-transformers
 Author: Phil Wang

{x_transformers-1.16.14.dist-info → x_transformers-1.16.15.dist-info}/RECORD

@@ -1,12 +1,12 @@
 x_transformers/__init__.py,sha256=FDb654rUx8FpXRd76B8q0diH8I7q-ZjTWEtEJ4UM21Y,701
-x_transformers/attend.py,sha256=Mm3KSsGKJ1PTqoCl0Za5K7S06Bs6cwaCMkC9-fqw1QY,10418
+x_transformers/attend.py,sha256=tDgL6ryV3QV2bU855iiyF19dbDWWdZ52dCPKk99sfOs,10382
 x_transformers/autoregressive_wrapper.py,sha256=u2celA8KeHm_Gd83Q7qaiLbJnwaDGdsbUck-JiokpKg,4446
 x_transformers/continuous_autoregressive_wrapper.py,sha256=pTiDqu6JRUlnQJQp_xHATYHy0lgSd6ERLqyiFO3pC-4,1575
 x_transformers/nonautoregressive_wrapper.py,sha256=AQLE4rA_Kh8VNoe9OzpwyeWson34sRkhks4dn4seNjI,10414
-x_transformers/x_transformers.py,sha256=n8NeIWo7WijFK4t9L5-ixlgvFnuxP-imEtW7T8e-pWg,54202
+x_transformers/x_transformers.py,sha256=rroq3Qq4XQ190XeVXeaiIsJdcYfidIOEt69N5DV_Rdo,54203
 x_transformers/xl_autoregressive_wrapper.py,sha256=-CAYjTtqrks8ZTxjYm2stOelZpU4MbZIvLjUxWO0P9Y,3988
-x_transformers-1.16.14.dist-info/LICENSE,sha256=As9u198X-U-vph5noInuUfqsAG2zX_oXPHDmdjwlPPY,1066
-x_transformers-1.16.14.dist-info/METADATA,sha256=lubdgZENatrU56P9BEsAvluo_yewrPydqf57coFFmZ0,666
-x_transformers-1.16.14.dist-info/WHEEL,sha256=pkctZYzUS4AYVn6dJ-7367OJZivF2e8RA9b_ZBjif18,92
-x_transformers-1.16.14.dist-info/top_level.txt,sha256=hO6KGpFuGucRNEtRfme4A_rGcM53AKwGP7RVlRIxS5Q,15
-x_transformers-1.16.14.dist-info/RECORD,,
+x_transformers-1.16.15.dist-info/LICENSE,sha256=As9u198X-U-vph5noInuUfqsAG2zX_oXPHDmdjwlPPY,1066
+x_transformers-1.16.15.dist-info/METADATA,sha256=-vQVqSIeWbDuvNGVV0JXOAyukJ4Ge6I0BHHcWhdTIdM,666
+x_transformers-1.16.15.dist-info/WHEEL,sha256=pkctZYzUS4AYVn6dJ-7367OJZivF2e8RA9b_ZBjif18,92
+x_transformers-1.16.15.dist-info/top_level.txt,sha256=hO6KGpFuGucRNEtRfme4A_rGcM53AKwGP7RVlRIxS5Q,15
+x_transformers-1.16.15.dist-info/RECORD,,