Update app.py

app.py

@@ -47,18 +47,22 @@ def analyze_input(image, question):
             pixel_values = preprocess_image(image)
         
         # Tokenize the question
-        input_ids = tokenizer(question, return_tensors="pt").input_ids.to(model.device)
-
+        tokenized = tokenizer(question, return_tensors="pt")
+        input_ids = tokenized.input_ids.to(model.device)
+        
+        # Calculate target size
+        tgt_size = input_ids.size(1) + 256  # Original input size + max new tokens
+        
         # Construct the model_inputs dictionary
         model_inputs = {
             "input_ids": input_ids,
-            "pixel_values": pixel_values
+            "pixel_values": pixel_values,
+            "tgt_sizes": [tgt_size]  # Add target sizes for generation
         }
 
         # Generate the response using the model
         outputs = model.generate(
             model_inputs=model_inputs,
-            max_new_tokens=256
         )
 
         # Decode the response