Update app.py

app.py

@@ -32,14 +32,13 @@ tokenizer = AutoTokenizer.from_pretrained(
     token=api_token
 )
 
-# Ensure eos_token_id is set
-if model.generation_config.eos_token_id is None:
-    eos_token_id = tokenizer.eos_token_id or tokenizer.pad_token_id
-    if eos_token_id is None:
-        raise ValueError(
-            "Both `eos_token_id` and `pad_token_id` are undefined in the tokenizer. Please define one explicitly."
-        )
-    model.generation_config.eos_token_id = eos_token_id
+# Ensure `eos_token_id` is properly set
+eos_token_id = tokenizer.eos_token_id or tokenizer.pad_token_id
+if eos_token_id is None:
+    raise ValueError(
+        "Neither `eos_token_id` nor `pad_token_id` is defined in the tokenizer. Please specify one explicitly."
+    )
+model.generation_config.eos_token_id = eos_token_id
 
 # Preprocess image
 def preprocess_image(image):
@@ -49,31 +48,31 @@ def preprocess_image(image):
 # Handle queries
 def analyze_input(image, question):
     try:
-        # Process the image if provided
+        # Prepare pixel values for image input
         pixel_values = None
         if image is not None:
             image = image.convert('RGB')
             pixel_values = preprocess_image(image)
-        
+
         # Tokenize the question
         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
-        
+
+        # Calculate target size for generation
+        tgt_size = input_ids.size(1) + 256  # Input size + max new tokens
+
         # Construct the model_inputs dictionary
         model_inputs = {
             "input_ids": input_ids,
             "pixel_values": pixel_values,
-            "tgt_sizes": [tgt_size]  # Add target sizes for generation
+            "tgt_sizes": [tgt_size]
         }
 
-        # Generate the response using the model
+        # Generate response
         outputs = model.generate(
-            model_inputs=model_inputs,
-            eos_token_id=model.generation_config.eos_token_id,  # Explicitly set eos_token_id
-            max_new_tokens=256
+            input_ids=model_inputs["input_ids"],
+            max_new_tokens=256,
+            eos_token_id=model.generation_config.eos_token_id
         )
 
         # Decode the response