nielsr HF staff commited on
Commit
87b061a
1 Parent(s): 95f7894

Add print statements

Browse files
Files changed (1) hide show
  1. modeling_cogvlm.py +7 -2
modeling_cogvlm.py CHANGED
@@ -434,12 +434,17 @@ class CogVLMModel(CogVLMPreTrainedModel):
434
  assert len(input_ids) == len(images), f"{len(input_ids)} {len(images)}"
435
  inputs_embeds = self.embed_tokens(input_ids)
436
 
437
- print("First values of text embeddings:", inputs_embeds[0, :3, :3])
438
-
439
  images_features = self.encode_images(images)
440
  images_features = rearrange(images_features, 'b n d -> (b n) d')
441
  images_features = images_features.to(dtype=inputs_embeds.dtype, device=inputs_embeds.device)
 
 
 
 
442
  inputs_embeds = inputs_embeds.index_put([token_type_ids == VISION_TOKEN_TYPE], images_features)
 
 
 
443
  else: # single-modality
444
  if token_type_ids is None:
445
  token_type_ids = torch.ones_like(input_ids, dtype=torch.long, device=input_ids.device) * LANGUAGE_TOKEN_TYPE
 
434
  assert len(input_ids) == len(images), f"{len(input_ids)} {len(images)}"
435
  inputs_embeds = self.embed_tokens(input_ids)
436
 
 
 
437
  images_features = self.encode_images(images)
438
  images_features = rearrange(images_features, 'b n d -> (b n) d')
439
  images_features = images_features.to(dtype=inputs_embeds.dtype, device=inputs_embeds.device)
440
+
441
+ print("First values of text embeddings:", inputs_embeds[0, :3, :3])
442
+ print("First values of images_features:", images_features[0, :3, :3])
443
+
444
  inputs_embeds = inputs_embeds.index_put([token_type_ids == VISION_TOKEN_TYPE], images_features)
445
+
446
+ print("First values of inputs_embeds after index_put:", inputs_embeds[0, :3, :3])
447
+
448
  else: # single-modality
449
  if token_type_ids is None:
450
  token_type_ids = torch.ones_like(input_ids, dtype=torch.long, device=input_ids.device) * LANGUAGE_TOKEN_TYPE