x-transformers 1.37.6__py3-none-any.whl → 1.37.8__py3-none-any.whl

x_transformers/attend.py

@@ -145,7 +145,6 @@ class Attend(Module):
 
         # l2 distance attention
 
-        assert not (flash and l2_distance), 'l2 distance attention does not work with flash attention just yet'
         self.l2_distance = l2_distance
 
         # add a key / value token composed of zeros
@@ -208,6 +207,17 @@ class Attend(Module):
         if v.ndim == 3:
             v = repeat(v, 'b ... -> b h ...', h = q.shape[1])
 
+        # handle maybe l2 distance
+
+        if self.l2_distance:
+            k_norm_sq = k.norm(dim = -1, keepdim = True) ** 2
+            k = F.pad(k, (0, 1), value = -1.)
+            k = torch.cat((k, k_norm_sq), dim = -1)
+
+            q_norm_sq = q.norm(dim = -1, keepdim = True) ** 2
+            q = torch.cat((2 * q, q_norm_sq), dim = -1)
+            q = F.pad(q, (0, 1), value = -1.)
+
         # handle scale - by default they scale by dim_head ** -0.5, but need to take care if using cosine sim attention
 
         if exists(self.scale):

{x_transformers-1.37.6.dist-info → x_transformers-1.37.8.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: x-transformers
-Version: 1.37.6
+Version: 1.37.8
 Summary: X-Transformers - Pytorch
 Home-page: https://github.com/lucidrains/x-transformers
 Author: Phil Wang

{x_transformers-1.37.6.dist-info → x_transformers-1.37.8.dist-info}/RECORD

@@ -1,5 +1,5 @@
 x_transformers/__init__.py,sha256=-MkQrSc37cTVDX7AOykxunYnqVtFlQ7lb0Cse5dsGWU,793
-x_transformers/attend.py,sha256=w3Gcy3gbeKGhIyKUiJKo8F90bPN-JcnQq3rjlVhG5sE,13155
+x_transformers/attend.py,sha256=we7mkwVCD7_ColUD8_Fj0HM5jjOaa3wbstllp_XXK4k,13434
 x_transformers/autoregressive_wrapper.py,sha256=DOJJCMMDOqDYKWy_IaG5IyKsXD3AW6amzfUgdAADOLY,10500
 x_transformers/continuous.py,sha256=cIVEdhfei258__ziV7kQBrJMxCel54bExBTDrO9rfCI,6450
 x_transformers/dpo.py,sha256=LjvWgCkqTl-UuehrzQ8nkX5guLr4whYwsmm7SKSwdls,3450
@@ -8,8 +8,8 @@ x_transformers/nonautoregressive_wrapper.py,sha256=2NU58hYMgn-4Jzg3mie-mXb0XH_dC
 x_transformers/x_transformers.py,sha256=-2fj6QcDSfMI5lJA_fzOW2mdzdS1C1LD6jMBtGQY48E,83752
 x_transformers/xl_autoregressive_wrapper.py,sha256=CvZMJ6A6PA-Y_bQAhnORwjJBSl6Vjq2IdW5KTdk8NI8,4195
 x_transformers/xval.py,sha256=QE1ltYZTR_eGgIHPP2BrMWVWVLqMW-OpDZh87BSmQEg,8563
-x_transformers-1.37.6.dist-info/LICENSE,sha256=As9u198X-U-vph5noInuUfqsAG2zX_oXPHDmdjwlPPY,1066
-x_transformers-1.37.6.dist-info/METADATA,sha256=IBUfeibj2CrLmh2-UPCokvC-ylmoBhc5OC-ZFRirvek,661
-x_transformers-1.37.6.dist-info/WHEEL,sha256=GV9aMThwP_4oNCtvEC2ec3qUYutgWeAzklro_0m4WJQ,91
-x_transformers-1.37.6.dist-info/top_level.txt,sha256=hO6KGpFuGucRNEtRfme4A_rGcM53AKwGP7RVlRIxS5Q,15
-x_transformers-1.37.6.dist-info/RECORD,,
+x_transformers-1.37.8.dist-info/LICENSE,sha256=As9u198X-U-vph5noInuUfqsAG2zX_oXPHDmdjwlPPY,1066
+x_transformers-1.37.8.dist-info/METADATA,sha256=fiT94VbrxWL-8jJBjxvFloWsH6n6reOGitRSlpAhvWs,661
+x_transformers-1.37.8.dist-info/WHEEL,sha256=GV9aMThwP_4oNCtvEC2ec3qUYutgWeAzklro_0m4WJQ,91
+x_transformers-1.37.8.dist-info/top_level.txt,sha256=hO6KGpFuGucRNEtRfme4A_rGcM53AKwGP7RVlRIxS5Q,15
+x_transformers-1.37.8.dist-info/RECORD,,