Update app.py

app.py

@@ -105,7 +105,7 @@ def preprocess(query: str) -> str:
     return prompt
 
 
-#@spaces.GPU
+@spaces.GPU
 def reply(message: str, history: list[str]) -> str:
     """
     Generates a response to the user’s message.
@@ -118,7 +118,7 @@ def reply(message: str, history: list[str]) -> str:
         str: The generated response from the language model.
     """
 
-    return llm(
+    yield llm(
         preprocess(message),
         max_new_tokens=512,
         return_full_text=False,