Update app.py

app.py

@@ -37,31 +37,31 @@ def preprocess_image(image):
     transform = ToTensor()
     return transform(image).unsqueeze(0).to(model.device)
 
+# Handle queries
 def analyze_input(image, question):
     try:
-        # Prepare inputs
-        if image:
-            # Process image
+        # Process the image if provided
+        pixel_values = None
+        if image is not None:
             image = image.convert('RGB')
             pixel_values = preprocess_image(image)
-            prompt = f"Given the medical image and question: {question}\nPlease provide a detailed analysis."
-            
-            # Model inputs for multimodal processing
-            model_inputs = {
-                "input_ids": tokenizer(prompt, return_tensors="pt").input_ids.to(model.device),
-                "pixel_values": pixel_values
-            }
-        else:
-            # Text-only processing
-            prompt = f"Medical question: {question}\nAnswer:"
-            model_inputs = {
-                "input_ids": tokenizer(prompt, return_tensors="pt").input_ids.to(model.device)
-            }
+        
+        # Tokenize the question
+        input_ids = tokenizer(question, return_tensors="pt").input_ids.to(model.device)
 
-        # Generate response using model's custom method
-        outputs = model.generate(**model_inputs, max_new_tokens=256)
+        # Construct the model_inputs dictionary
+        model_inputs = {
+            "input_ids": input_ids,
+            "pixel_values": pixel_values
+        }
 
-        # Decode and clean response
+        # Generate the response using the model
+        outputs = model.generate(
+            model_inputs=model_inputs,
+            max_new_tokens=256
+        )
+
+        # Decode the response
         response = tokenizer.decode(outputs[0], skip_special_tokens=True)
         return {"status": "success", "response": response}