🐛

dalle2_pytorch/dalle2_pytorch.py

@@ -647,9 +647,12 @@ class DiffusionPrior(BaseGaussianDiffusion):
         )
 
         if exists(clip):
-            assert isinstance(clip, CLIP)
+            if isinstance(clip, CLIP):
+                clip = XClipAdapter(clip)
+
+            assert isinstance(clip, BaseClipAdapter)
             freeze_model_and_make_eval_(clip)
-            self.clip = XClipAdapter(clip)
+            self.clip = clip
         else:
             assert exists(image_embed_dim), 'latent dimension must be given, if training prior network without CLIP given'
             self.clip = None
@@ -740,7 +743,7 @@ class DiffusionPrior(BaseGaussianDiffusion):
         text_cond = dict(text_embed = text_embed)
 
         if self.condition_on_text_encodings:
-            text_cond = {**text_cond, 'text_encodings': text_encodings, 'mask': text_mask}
+            text_cond = {**text_cond, 'text_encodings': text_encodings, 'mask': text != 0}
 
         image_embeds = self.p_sample_loop((batch_size, image_embed_dim), text_cond = text_cond)
         text_embeds = text_cond['text_embed']
@@ -1248,6 +1251,8 @@ class Decoder(BaseGaussianDiffusion):
             clip = XClipAdapter(clip)
 
         freeze_model_and_make_eval_(clip)
+        assert isinstance(clip, BaseClipAdapter)
+
         self.clip = clip
         self.clip_image_size = clip.image_size
         self.channels = clip.image_channels

setup.py

@@ -10,7 +10,7 @@ setup(
       'dream = dalle2_pytorch.cli:dream'
     ],
   },
-  version = '0.0.60',
+  version = '0.0.62',
   license='MIT',
   description = 'DALL-E 2',
   author = 'Phil Wang',