Update app.py

app.py

@@ -433,7 +433,7 @@ async def handle_chat(chat_request: ChatRequest, token: str = Cookie(None)):
                 model="",
                 messages=History[i],
             )
-        History[i].append({"role": "assistant", "content": chat_request.prompt})
+        History[i].append({"role": "assistant", "content": stream.choices[0].message.content})
         return {"text": stream.choices[0].message.content,
                "time": int((stream.timings["prompt_ms"] + stream.timings["predicted_ms"])/1000 - 3),
                "t_per_sec": round(stream.timings["predicted_per_second"] + 0.2, 2)}