Update app.py

app.py

@@ -31,7 +31,7 @@ def chat(message, history):
         if item[1] is not None:
             chat.append({"role": "assistant", "content": item[1]})
     chat.append({"role": "user", "content": message})
-    messages = tok.apply_chat_template(chat, tokenize=False)
+    messages = tok.apply_chat_template(chat, tokenize=False, add_generation_prompt=True)
     # Tokenize the messages string
     model_inputs = tok([messages], return_tensors="pt").to(device)
     streamer = TextIteratorStreamer(