x-transformers 2.1.11__py3-none-any.whl → 2.1.14__py3-none-any.whl

x_transformers/belief_state_wrapper.py

@@ -132,11 +132,11 @@ class BeliefStateWrapper(Module):
 
         # get the encoded suffix token once
 
-        if not exists(suffix):
-            suffix = out[:, 0:0]
+        if exists(suffix):
+            if suffix.ndim == 1:
+                suffix = repeat(suffix, 'n -> b n', b = batch)
 
-        if suffix.ndim == 1:
-            suffix = repeat(suffix, 'n -> b n', b = batch)
+            suffix = suffix.flip(1) # reverse autoregressive
 
         suffix_sos_tokens = rearrange(self.suffix_token, 'd -> 1 1 d')
 
@@ -148,6 +148,10 @@ class BeliefStateWrapper(Module):
             return_embeddings = True
         )
 
+        # pick out the last embedding for fill in the model
+
+        suffix_embed = suffix_embed[:, -1:]
+
         # sampling up to seq_len
 
         for _ in range(seq_len):

x_transformers/x_transformers.py

@@ -2898,6 +2898,15 @@ class TransformerWrapper(Module):
         to_logits_kwargs = dict(),
         **kwargs,
     ):
+
+        # if sequence is None, auto create an empty one if `prepend_embeds` was supplied
+
+        if not exists(x):
+            assert exists(prepend_embeds)
+            x = prepend_embeds.new_empty((prepend_embeds.shape[0], 0), dtype = torch.long)
+
+        # shapes and variables
+
         b, n, device, num_mems, has_memory_tokens, emb_frac_gradient, orig_mask = x.shape[0], x.shape[1], x.device, self.num_memory_tokens, self.num_memory_tokens > 0, self.emb_frac_gradient, mask
 
         return_hiddens = return_mems | return_attn | return_intermediates | return_attn_z_loss

{x_transformers-2.1.11.dist-info → x_transformers-2.1.14.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: x-transformers
-Version: 2.1.11
+Version: 2.1.14
 Summary: X-Transformers
 Project-URL: Homepage, https://pypi.org/project/x-transformers/
 Project-URL: Repository, https://github.com/lucidrains/x-transformers

{x_transformers-2.1.11.dist-info → x_transformers-2.1.14.dist-info}/RECORD

@@ -1,16 +1,16 @@
 x_transformers/__init__.py,sha256=NDoiBivau559WQ0FvXG4ssU3Il9aoHmTIUFN_1juz0s,911
 x_transformers/attend.py,sha256=-5BWWhFsp7tvZTdN91Ay5SqOjyj9uOs-122vFvoO6b4,17253
 x_transformers/autoregressive_wrapper.py,sha256=reLCno9Z9pchVU79tBF8OMo21LwSZ67KAeB83jqkyAc,10505
-x_transformers/belief_state_wrapper.py,sha256=mpfTNZb8gadbtlpG2TxyfIMWkMVM4vigFDqCJ_mjxSU,8711
+x_transformers/belief_state_wrapper.py,sha256=qkVg_Nd77G09rjgrPJJkYN7NE58Rxl2DQUR475Zk4vA,8849
 x_transformers/continuous.py,sha256=p0sCAiH1na236ygwgL1Yyhu36eZBf9cZvoW1JyP_fFE,7073
 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=-80N4sqUr3sR51Ms4wCfc4jhxnPwf0ApNR4xfIsasfQ,110142
+x_transformers/x_transformers.py,sha256=fqgtIs6__JpLWMnJa8AY5OW3AJ2GR1B5p-9TsWdiOIU,110425
 x_transformers/xl_autoregressive_wrapper.py,sha256=CvZMJ6A6PA-Y_bQAhnORwjJBSl6Vjq2IdW5KTdk8NI8,4195
 x_transformers/xval.py,sha256=7S00kCuab4tWQa-vf-z-XfzADjVj48MoFIr7VSIvttg,8575
-x_transformers-2.1.11.dist-info/METADATA,sha256=miwxNJVS0ZNJlw3qeJaGaxunTRxLgmz9WITWi_jnXcc,87571
-x_transformers-2.1.11.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-x_transformers-2.1.11.dist-info/licenses/LICENSE,sha256=As9u198X-U-vph5noInuUfqsAG2zX_oXPHDmdjwlPPY,1066
-x_transformers-2.1.11.dist-info/RECORD,,
+x_transformers-2.1.14.dist-info/METADATA,sha256=WIAtwwv4SrEpN239S56A2MVxTFKxo4Csyj4VJBrUee0,87571
+x_transformers-2.1.14.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+x_transformers-2.1.14.dist-info/licenses/LICENSE,sha256=As9u198X-U-vph5noInuUfqsAG2zX_oXPHDmdjwlPPY,1066
+x_transformers-2.1.14.dist-info/RECORD,,