Xinference Chat Bot 每次对话多轮就会卡死

[andylzming]

Describe the bug

Xinference Chat Bot 每次对话多轮（一般两三轮）就会卡死，详见截图。

To Reproduce

To help us to reproduce this bug, please provide information below:

1. Python ：3.10.6
2. xinference : 0.9.4
3. Versions of crucial packages.
4. Full stack of the error.
5. Minimized code to reproduce the error.

Expected behavior

A clear and concise description of what you expected to happen.

Additional context

Add any other context about the problem here.

[ChengjieLi28]

@andylzming 。我用一样的模型可以复现此问题（不一定百分百，我换个模型有时不会触发），gradio 版本：

gradio                        3.50.1
gradio_client                 0.6.1

打开F12可以看到console上有报错，然后网络中的ws中其实模型的回答已经传回来了，只是gradio没显示出来。猜测gradio版本有问题。
详见：gradio-app/gradio#6613
和 gradio-app/gradio#3943

按照issue里面，gradio降级到3.41，我就再也没出现这样的问题，你可以试下。

[andylzming]

@ChengjieLi28

gradio降级到3.41，submit 按钮点击无效。

以下两个版本都会出现对话卡死现象

(xinference) [root@gpu-server gradio]# pip list | grep gradio
gradio                        3.47.1
gradio_client                 0.6.0

(xinference) [root@gpu-server depends]# ll xinference-dependences/ | grep gradio
-rw-r--r--. 1 root root  20298198 12月 19 21:55 gradio-3.50.2-py3-none-any.whl
-rw-r--r--. 1 root root    299220 12月 19 21:55 gradio_client-0.6.1-py3-none-any.whl

控制台如下：

[andylzming]

qwen-14b 模型对话多轮正常，chatglm3-6b 不行。
另外，通过 dify 使用 xinference 与 chatglm3-6b 通信报以下错误：

• 错误日志

INFO 04-10 16:30:35 llm_engine.py:653] Avg prompt throughput: 24.8 tokens/s, Avg generation throughput: 6.5 tokens/s, Running: 1 reqs, Swapped: 0 reqs, Pending: 0 reqs, GPU KV cache usage: 0.0%, CPU KV cache usage: 0.0%
INFO 04-10 16:30:35 async_llm_engine.py:111] Finished request 2f3a1d40-f779-11ee-b1b4-80615f20f615.
2024-04-10 16:30:35,419 xinference.api.restful_api 27390 ERROR    [address=127.0.0.1:34773, pid=24418] 0
Traceback (most recent call last):
  File "/home/miniconda3/envs/xinference/lib/python3.10/site-packages/xinference/api/restful_api.py", line 1394, in create_chat_completion
    data = await model.chat(prompt, system_prompt, chat_history, kwargs)
  File "/home/miniconda3/envs/xinference/lib/python3.10/site-packages/xoscar/backends/context.py", line 227, in send
    return self._process_result_message(result)
  File "/home/miniconda3/envs/xinference/lib/python3.10/site-packages/xoscar/backends/context.py", line 102, in _process_result_message
    raise message.as_instanceof_cause()
  File "/home/miniconda3/envs/xinference/lib/python3.10/site-packages/xoscar/backends/pool.py", line 659, in send
    result = await self._run_coro(message.message_id, coro)
  File "/home/miniconda3/envs/xinference/lib/python3.10/site-packages/xoscar/backends/pool.py", line 370, in _run_coro
    return await coro
  File "/home/miniconda3/envs/xinference/lib/python3.10/site-packages/xoscar/api.py", line 384, in __on_receive__
    return await super().__on_receive__(message)  # type: ignore
  File "xoscar/core.pyx", line 558, in __on_receive__
    raise ex
  File "xoscar/core.pyx", line 520, in xoscar.core._BaseActor.__on_receive__
    async with self._lock:
  File "xoscar/core.pyx", line 521, in xoscar.core._BaseActor.__on_receive__
    with debug_async_timeout('actor_lock_timeout',
  File "xoscar/core.pyx", line 526, in xoscar.core._BaseActor.__on_receive__
    result = await result
  File "/home/miniconda3/envs/xinference/lib/python3.10/site-packages/xinference/core/utils.py", line 45, in wrapped
    ret = await func(*args, **kwargs)
  File "/home/miniconda3/envs/xinference/lib/python3.10/site-packages/xinference/core/model.py", line 79, in wrapped_func
    ret = await fn(self, *args, **kwargs)
  File "/home/miniconda3/envs/xinference/lib/python3.10/site-packages/xoscar/api.py", line 462, in _wrapper
    r = await func(self, *args, **kwargs)
  File "/home/miniconda3/envs/xinference/lib/python3.10/site-packages/xinference/core/model.py", line 375, in chat
    response = await self._call_wrapper(
  File "/home/miniconda3/envs/xinference/lib/python3.10/site-packages/xinference/core/model.py", line 103, in _async_wrapper
    return await fn(*args, **kwargs)
  File "/home/miniconda3/envs/xinference/lib/python3.10/site-packages/xinference/core/model.py", line 325, in _call_wrapper
    ret = await fn(*args, **kwargs)
  File "/home/miniconda3/envs/xinference/lib/python3.10/site-packages/xinference/model/llm/vllm/core.py", line 439, in async_chat
    return self._tool_calls_completion(
  File "/home/miniconda3/envs/xinference/lib/python3.10/site-packages/xinference/model/llm/utils.py", line 601, in _tool_calls_completion
    content, func, args = cls._eval_chatglm3_arguments(c, tools)
  File "/home/miniconda3/envs/xinference/lib/python3.10/site-packages/xinference/model/llm/utils.py", line 548, in _eval_chatglm3_arguments
    if isinstance(c[0], str):
KeyError: [address=127.0.0.1:34773, pid=24418] 0

[github-actions]

This issue is stale because it has been open for 7 days with no activity.

[github-actions]

This issue was closed because it has been inactive for 5 days since being marked as stale.

[Matrix0816]

我也遇到模型多轮对话后崩掉的问题，未找到解决方案Qwen 14B