vidhanm commited on
Commit
3253deb
·
1 Parent(s): a4644a0

trying to fix typerror

Browse files
Files changed (1) hide show
  1. app.py +7 -6
app.py CHANGED
@@ -31,8 +31,7 @@ print(f"Using device: {device}")
31
  # --- Configuration for model components ---
32
  model_id_for_weights = "lusxvr/nanoVLM-222M"
33
  image_processor_id = "openai/clip-vit-base-patch32"
34
- # Load the tokenizer from its original source to ensure all files are present
35
- tokenizer_id = "gpt2" # Changed from "lusxvr/nanoVLM-222M"
36
 
37
  image_processor = None
38
  tokenizer = None
@@ -41,20 +40,21 @@ model = None
41
  if VisionLanguageModel:
42
  try:
43
  print(f"Attempting to load CLIPImageProcessor from: {image_processor_id}")
 
44
  image_processor = CLIPImageProcessor.from_pretrained(image_processor_id, trust_remote_code=True)
45
  print("CLIPImageProcessor loaded.")
46
 
47
  print(f"Attempting to load GPT2TokenizerFast from: {tokenizer_id}")
48
  tokenizer = GPT2TokenizerFast.from_pretrained(tokenizer_id, trust_remote_code=True)
49
  if tokenizer.pad_token is None:
50
- tokenizer.pad_token = tokenizer.eos_token # Important for GPT-2
51
  print("Set tokenizer pad_token to eos_token.")
52
  print("GPT2TokenizerFast loaded.")
53
 
54
  print(f"Attempting to load model weights from {model_id_for_weights} using VisionLanguageModel.from_pretrained")
 
55
  model = VisionLanguageModel.from_pretrained(
56
- model_id_for_weights,
57
- trust_remote_code=True
58
  ).to(device)
59
  print("Model loaded successfully.")
60
  model.eval()
@@ -69,6 +69,7 @@ if VisionLanguageModel:
69
  else:
70
  print("Custom VisionLanguageModel class not imported, cannot load model.")
71
 
 
72
  def prepare_inputs(text_list, image_input, image_processor_instance, tokenizer_instance, device_to_use):
73
  if image_processor_instance is None or tokenizer_instance is None:
74
  raise ValueError("Image processor or tokenizer not initialized.")
@@ -152,7 +153,7 @@ iface = gr.Interface(
152
  [example_image_url, "a photo of a"],
153
  [example_image_url, "Describe the image in detail."],
154
  ],
155
- # cache_examples=True, # Temporarily commented out to ensure Gradio starts with minimal config
156
  allow_flagging="never"
157
  )
158
 
 
31
  # --- Configuration for model components ---
32
  model_id_for_weights = "lusxvr/nanoVLM-222M"
33
  image_processor_id = "openai/clip-vit-base-patch32"
34
+ tokenizer_id = "gpt2"
 
35
 
36
  image_processor = None
37
  tokenizer = None
 
40
  if VisionLanguageModel:
41
  try:
42
  print(f"Attempting to load CLIPImageProcessor from: {image_processor_id}")
43
+ # trust_remote_code for HF's classes is fine if they support it.
44
  image_processor = CLIPImageProcessor.from_pretrained(image_processor_id, trust_remote_code=True)
45
  print("CLIPImageProcessor loaded.")
46
 
47
  print(f"Attempting to load GPT2TokenizerFast from: {tokenizer_id}")
48
  tokenizer = GPT2TokenizerFast.from_pretrained(tokenizer_id, trust_remote_code=True)
49
  if tokenizer.pad_token is None:
50
+ tokenizer.pad_token = tokenizer.eos_token
51
  print("Set tokenizer pad_token to eos_token.")
52
  print("GPT2TokenizerFast loaded.")
53
 
54
  print(f"Attempting to load model weights from {model_id_for_weights} using VisionLanguageModel.from_pretrained")
55
+ # Removed trust_remote_code=True as the custom VisionLanguageModel.from_pretrained doesn't expect it.
56
  model = VisionLanguageModel.from_pretrained(
57
+ model_id_for_weights
 
58
  ).to(device)
59
  print("Model loaded successfully.")
60
  model.eval()
 
69
  else:
70
  print("Custom VisionLanguageModel class not imported, cannot load model.")
71
 
72
+ # ... (rest of the app.py remains the same) ...
73
  def prepare_inputs(text_list, image_input, image_processor_instance, tokenizer_instance, device_to_use):
74
  if image_processor_instance is None or tokenizer_instance is None:
75
  raise ValueError("Image processor or tokenizer not initialized.")
 
153
  [example_image_url, "a photo of a"],
154
  [example_image_url, "Describe the image in detail."],
155
  ],
156
+ # cache_examples=True, # Keep commented out for now
157
  allow_flagging="never"
158
  )
159