Add print statements

modeling_cogvlm.py

@@ -434,12 +434,17 @@ class CogVLMModel(CogVLMPreTrainedModel):
                 assert len(input_ids) == len(images), f"{len(input_ids)} {len(images)}"
                 inputs_embeds = self.embed_tokens(input_ids)
 
-                print("First values of text embeddings:", inputs_embeds[0, :3, :3])
-
                 images_features = self.encode_images(images)
                 images_features = rearrange(images_features, 'b n d -> (b n) d')
                 images_features = images_features.to(dtype=inputs_embeds.dtype, device=inputs_embeds.device)
+
+                print("First values of text embeddings:", inputs_embeds[0, :3, :3])
+                print("First values of images_features:", images_features[0, :3, :3])
+
                 inputs_embeds = inputs_embeds.index_put([token_type_ids == VISION_TOKEN_TYPE], images_features)
+
+                print("First values of inputs_embeds after index_put:", inputs_embeds[0, :3, :3])
+
             else:  # single-modality
                 if token_type_ids is None:
                     token_type_ids = torch.ones_like(input_ids, dtype=torch.long, device=input_ids.device) * LANGUAGE_TOKEN_TYPE