Update app.py

app.py

@@ -40,7 +40,7 @@ def get_reward_score(prompt: str, response: str) -> float:
         # Use the same chat template as used during model training.
         messages = [{"role": "user", "content": prompt}, {"role": "assistant", "content": response}]
         # The model will handle moving inputs to the correct device automatically.
-        input_ids = tokenizer.apply_chat_template(messages, return_tensors="pt") # <<< REMOVED .to(DEVICE)
+        input_ids = tokenizer.apply_chat_template(messages, return_tensors="pt").to(DEVICE) # <<< REMOVED .to(DEVICE)
 
         with torch.no_grad():
             score = model(input_ids).logits.item()