Update modeling_gpt2vision.py

modeling_gpt2vision.py

@@ -49,7 +49,7 @@ class GPT2Vision(PreTrainedModel):
         self.tokenizer = tokenizer
         tokenizer.pad_token = tokenizer.eos_token
         self.image_token_id = self.tokenizer.convert_tokens_to_ids(IMAGE_TOKEN)
-        self.img_tokens = 197  # This should match IMG_TOKENS in your training code
+        self.img_tokens = 197  
 
     @property
     def device(self):
@@ -61,12 +61,11 @@ class GPT2Vision(PreTrainedModel):
         if isinstance(text, str):
             text = [text]
 
-        print("text",text)
         text_inputs = self.tokenizer(
             text,
             padding='max_length',
             truncation=True,
-            max_length=768,
+            max_length=384,
             return_tensors="pt",
         ).to(device)
         
@@ -97,15 +96,6 @@ class GPT2Vision(PreTrainedModel):
 
         return inputs_embeds, attention_mask
 
-    def forward(self, batch, **kwargs):
-        inputs_embeds, attention_mask = self.preprocess_inputs(batch)
-    
-        outputs = self.language_model(
-            inputs_embeds=inputs_embeds,
-            attention_mask=attention_mask,
-            **kwargs
-        )
-        return outputs
         
     def generate(self, question, image, max_new_tokens=30, **kwargs):
         prompt = f"\n\nQuestion: {question}\n\nAnswer:"