InfoGPT - Make large language models such as ChatGPT/Wenxin Yiyan easier to use

             

InfoGPT is ChatGPTa model developed by the author in the process of learning and other large language models 开源软件. As of version v0.0.14, it supports three platforms: openai api proxy, telegram robot, and WeChat official account. WeChat official account integration is more troublesome, so I will make a special introduction.

Project address: https://github.com/alwqx/infogpt

Welcome everyone to try and raise demand

for individual users

1. OpenAI API delegate

2. Shared API Key

Telegram Bot

binary run

1. Go to the releases ^[1] webpage to download the latest version of the compressed package. Here we take the Apple M2 chip compressed package infogpt-v0.0.14-darwin-arm64.tar.gzas an example .

$ tar xzvf infogpt-v0.0.14-darwin-arm64.tar.gz
x README.md
x configs/
x configs/config.yaml
x infogpt

1. Modify the configuration fileconfigs/config.yaml

server:
  http:
    addr: 0.0.0.0:6060
    timeout: 300s
  grpc:
    addr: 0.0.0.0:6061
    timeout: 2s
admin:
  openai_api_key: ""
  proxy_url: ""
  gin_ratelimit: "20-D"
  telegram:
    token: ""
    # 机器人的请求限制，原则上应该比 user_ratelimit 大，为空则不进行限制
    ratelimit: ""
    # 每个用户的请求限制，为空则不进行限制
    user_ratelimit: "20-D"
    exclude_keys: ["foo", "bar"]
  wechat:
    app_id: ""
    app_secret: ""
    token: ""
    encoding_aes_key: ""
    # 机器人的请求限制，原则上应该比 user_ratelimit 大，为空则不进行限制
    ratelimit: ""
    # 每个用户的请求限制，为空则不进行限制
    user_ratelimit: "20-D"
    exclude_keys: ["foo", "bar"]
    auto_replay:
      "001": "infogpt infogpt1"

1. run the binary

# geek @ geekdeMBP in ~/Downloads/infogptlab [22:18:36]
$ ./infogpt -conf configs
2023/04/18 22:18:54 maxprocs: Leaving GOMAXPROCS=12: CPU quota undefined
DEBUG msg=config loaded: config.yaml format: yaml
INFO module=service/admin ts=2023-04-18T22:18:54+08:00 caller=service.go:71 service.id=geekdeMBP.lan service.name= service.version= trace.id= span.id= msg=[NewAdminService] no proxy_url, use http.DefaultClient
WARN msg=[NewAdminService] not enable telegram bot, skip

[GIN-debug] GET    /hello                    --> infogpt/internal/server.ginHello (5 handlers)
[GIN-debug] GET    /openaiproxy/*path        --> infogpt/internal/server.openaiProxy (5 handlers)
WARN msg=not enable wechat, skip

Note : By default, there is only the proxy function, that is, the request sent to infogpt is forwarded to OpenAI's API server.

1. Background process

nohup ./infogpt -conf configs > infogpt.log 2>&1 &

1. request a demo

$ curl --location 'localhost:6060/openaiproxy/v1/completions' \
--header 'Authorization: Bearer YOUR_TOKEN' \
--header 'Content-Type: application/json' \
--data '{
    "model": "text-davinci-003",
    "prompt": "你好",
    "max_tokens": 1024,
    "temperature": 0
  }'
{
  "id": "cmpl-76grTMyULgpL8MT8iAyqR9ZChwyqX",
  "object": "text_completion",
  "created": 1681829083,
  "model": "text-davinci-003",
  "choices": [
    {
      "text": "\n\n 你好！",
      "index": 0,
      "logprobs": null,
      "finish_reason": "stop"
    }
  ],
  "usage": {
    "prompt_tokens": 4,
    "completion_tokens": 9,
    "total_tokens": 13
  }
}

For details, refer to the document deploy ^[2]

Docker run

docker run -d --name infogpt --rm -p 6060:6060 -p 6061:6061 -v </path/to/your/configs>:/data/conf infogpt:latest

All

• [x] Support OpenAI Proxy

• [x] REST/gRPC interface

  • [x] chat

  • [x] Generate web article summaries

  • [x] Generate book content summaries

• [x] Telegram

  • [x] /chat /url /book three commands

  • [x] use frequency limit

• [x] WeChat official account limited support

  • [x] use frequency limit

  • [x] Timeout caching Q&A

• [ ] Support ChatGPT Proxy

• [ ] Slack

• [ ] Generate a summary of the contents of the file

thank you

The development of this open source project is inseparable from the support of these open source projects:

• go-kratos^[3]

References

[1]

releases: https://github.com/alwqx/infogpt/releases

[2]

deploy: docs/deploy.md

[3]

go-kratos: https://github.com/go-kratos/kratos