Update model.py

model.py

@@ -26,7 +26,7 @@ def run(message: str,
         history.append({"role": "assistant", "content": i[1]})
     print(history)
     history.append({"role": "user", "content": message})
-    for response in llm.create_chat_completion(history,stream=True,max_tokens=-1,temperature=temperature,top_k=top_k,top_p=top_p,repeat_penalty=1.1):
+    for response in llm.create_chat_completion(history,stop=["</s>"],stream=True,max_tokens=-1,temperature=temperature,top_k=top_k,top_p=top_p,repeat_penalty=1.1):
         if "content" in response["choices"][0]["delta"]:
             result = result + response["choices"][0]["delta"]["content"]
             yield result