[Bug] InternLM 模型重复回答 & 后端 json 解析持续报错

[crazysteeaam]

环境：使用lmdeploy internlm 部署
和教程基本一致

反馈两个bug：

1. 前端一次请求，但后端死循环持续请求模型API，没有跳出循环，只能手动停止。

问题一初步判断是 InternLM 模型返回结果是 "finish_reason": null，因此永远不会进入这个分支。
openaoe/backend/service/service_internlm.py

if choice['finish_reason']:
    stop_flag = True

OpenAI API 返回的是 "finish_reason": stop，就可以正常跳出。

2. 后端 json 解析持续报错，同时在疯狂请求模型API
   AOE后端命令行：

INFO:     127.0.0.1:50810 - "POST /v1/internlm/v1/chat/completions HTTP/1.1" 200 OK
JSON Parse Error: Expecting value: line 1 column 2 (char 1)
JSON Parse Error: Expecting value: line 1 column 1 (char 0)
JSON Parse Error: Expecting value: line 1 column 2 (char 1)
JSON Parse Error: Expecting value: line 1 column 1 (char 0)
JSON Parse Error: Expecting value: line 1 column 2 (char 1)
JSON Parse Error: Expecting value: line 1 column 1 (char 0)
JSON Parse Error: Expecting value: line 1 column 2 (char 1)
......

模型服务命令行：

INFO:     127.0.0.1:36372 - "POST /v1/chat/completions HTTP/1.1" 200 OK
INFO:     127.0.0.1:36382 - "POST /v1/chat/completions HTTP/1.1" 200 OK
INFO:     127.0.0.1:36394 - "POST /v1/chat/completions HTTP/1.1" 200 OK
INFO:     127.0.0.1:36396 - "POST /v1/chat/completions HTTP/1.1" 200 OK
INFO:     127.0.0.1:36400 - "POST /v1/chat/completions HTTP/1.1" 200 OK
INFO:     127.0.0.1:36406 - "POST /v1/chat/completions HTTP/1.1" 200 OK
INFO:     127.0.0.1:36422 - "POST /v1/chat/completions HTTP/1.1" 200 OK
INFO:     127.0.0.1:36426 - "POST /v1/chat/completions HTTP/1.1" 200 OK
INFO:     127.0.0.1:36428 - "POST /v1/chat/completions HTTP/1.1" 200 OK
INFO:     127.0.0.1:36434 - "POST /v1/chat/completions HTTP/1.1" 200 OK
......

[JiaYingLii]

Thanks for your post, we add a retry times parameter to limit the forever retry.