x-transformers 1.43.2__py3-none-any.whl → 1.43.5__py3-none-any.whl

x_transformers/x_transformers.py

@@ -38,6 +38,7 @@ class LayerIntermediates:
     attn_z_loss:        Tensor | None = None
     mems:               Tensor | None = None
     memory_tokens:      Tensor | None = None
+    logit_entropies:    Tensor | None = None
 
 LinearNoBias = partial(nn.Linear, bias = False)
 
@@ -136,6 +137,15 @@ def or_reduce(masks):
         head = head | rest
     return head
 
+# entropy
+
+def calc_entropy(
+    t: Tensor,
+    is_prob = False
+):
+    prob = t.softmax(dim = -1) if not is_prob else t
+    return -(prob * log(prob)).sum(dim = -1)
+
 # auxiliary loss helpers
 
 def calc_z_loss(
@@ -2270,16 +2280,16 @@ class AttentionLayers(Module):
         if self.need_condition:
             final_norm = maybe(partial)(final_norm, **norm_kwargs)
 
-        if self.resi_dual:
-            x = x + final_norm(outer_residual)
-        else:
-            x = final_norm(x)
-
         # take care of multistreams if needed, use sum for now
 
         if is_multistream:
             x = reduce(x, '(b s) n d -> b n d', 'sum', s = streams)
 
+        if self.resi_dual:
+            x = x + final_norm(outer_residual)
+        else:
+            x = final_norm(x)
+
         if not return_hiddens:
             return x
 
@@ -2592,6 +2602,7 @@ class TransformerWrapper(Module):
         return_embeddings = False,
         return_logits_and_embeddings = False,
         return_intermediates = False,
+        return_logit_entropies = False,
         mask = None,
         return_mems = False,
         return_attn = False,
@@ -2809,6 +2820,12 @@ class TransformerWrapper(Module):
         else:
             out = logits
 
+        # logit entropies
+
+        if return_logit_entropies:
+            intermediates.logit_entropies = calc_entropy(logits)
+            return_intermediates = True
+
         # aux loss
 
         if return_attn_z_loss:

{x_transformers-1.43.2.dist-info → x_transformers-1.43.5.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: x-transformers
-Version: 1.43.2
+Version: 1.43.5
 Summary: X-Transformers - Pytorch
 Home-page: https://github.com/lucidrains/x-transformers
 Author: Phil Wang

{x_transformers-1.43.2.dist-info → x_transformers-1.43.5.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=SsmQacZJ95oTBCcu33m42-_xhBlnoO4G9MkGRQNF_rI,100392
+x_transformers/x_transformers.py,sha256=z3RH6jvjcxaAVfZoCS0HWrE0Gy55-eXOKtzRt7rRRIw,100811
 x_transformers/xl_autoregressive_wrapper.py,sha256=CvZMJ6A6PA-Y_bQAhnORwjJBSl6Vjq2IdW5KTdk8NI8,4195
 x_transformers/xval.py,sha256=7S00kCuab4tWQa-vf-z-XfzADjVj48MoFIr7VSIvttg,8575
-x_transformers-1.43.2.dist-info/LICENSE,sha256=As9u198X-U-vph5noInuUfqsAG2zX_oXPHDmdjwlPPY,1066
-x_transformers-1.43.2.dist-info/METADATA,sha256=ilN5k-uCKnN1LFgV_mTKt6zX2f1NxrUF7-nqwR_yJ4c,738
-x_transformers-1.43.2.dist-info/WHEEL,sha256=PZUExdf71Ui_so67QXpySuHtCi3-J3wvF4ORK6k_S8U,91
-x_transformers-1.43.2.dist-info/top_level.txt,sha256=hO6KGpFuGucRNEtRfme4A_rGcM53AKwGP7RVlRIxS5Q,15
-x_transformers-1.43.2.dist-info/RECORD,,
+x_transformers-1.43.5.dist-info/LICENSE,sha256=As9u198X-U-vph5noInuUfqsAG2zX_oXPHDmdjwlPPY,1066
+x_transformers-1.43.5.dist-info/METADATA,sha256=crd2xA3NbodKbOz9xY4D1j3XDbTmaY9vwkZZJOGoEw4,738
+x_transformers-1.43.5.dist-info/WHEEL,sha256=PZUExdf71Ui_so67QXpySuHtCi3-J3wvF4ORK6k_S8U,91
+x_transformers-1.43.5.dist-info/top_level.txt,sha256=hO6KGpFuGucRNEtRfme4A_rGcM53AKwGP7RVlRIxS5Q,15
+x_transformers-1.43.5.dist-info/RECORD,,