Skip to content

Latest commit

 

History

History
111 lines (92 loc) · 4.36 KB

README.md

File metadata and controls

111 lines (92 loc) · 4.36 KB
Raw

Take chatGPT into command line.

stream

Setup

  1. clone this repo
  2. pip3 install -U -r requirements.txt
  3. copy demo_config.json to config.json
  4. get your OPENAI_API_KEY and put it in config.json

Run

$ ./gptcli.py -h
usage: gptcli.py [-h] [-c CONFIG]

options:
  -h, --help  show this help message and exit
  -c CONFIG   path to your config.json (default: config.json)

Sample config.json:

{
    "api_key": "sk-xxx",
    "api_base": "https://chat.pppan.net/v1",
    "model": "gpt-3.5-turbo",
    "context": 2,
    "stream": true,
    "stream_render": true,
    "showtokens": false,
    "proxy": "socks5://localhost:1080",
    "prompt": [
        { "role": "system", "content": "If your response contains code, show with syntax highlight, for example ```js\ncode\n```" }
    ]
}
  • (required) api_key: OpenAI's api key. will read from OPENAI_API_KEY envronment variable if not set
  • (optional) api_base: OpenAI's api base url. Can set to a server reverse proxy, for example Azure OpenAI Service or chatgptProxyAPI. By default it's from OPENAI_API_BASE or just https://api.openai.com/v1;
  • (optional) api_type: OpenAI's api type, read from env OPENAI_API_TYPE by default;
  • (optional) api_version: OpenAI's api version, read from env OPENAI_API_VERSION by default;
  • (optional) api_organization: OpenAI's organization info, read from env OPENAI_ORGANIZATION by default;
  • (optional) model: OpenAI's chat model, by default it's gpt-3.5-turbo; choices are:
    • gpt-3.5-turbo
    • gpt-4
    • gpt-4-32k
  • (optional) context: Chat session context, choices are:
    • 0: no context provided for every chat request, cost least tokens, but AI don't kown what you said before;
    • 1: only use previous user questions as context;
    • 2: use both previous questions and answers as context, would cost more tokens;
  • (optional) stream: Output in stream mode;
  • (optional) stream_render: Render markdown in stream mode, you can disable it to avoid some UI bugs;
  • (optional) showtokens: Print used tokens after every chat;
  • (optional) proxy: Use http/https/socks4a/socks5 proxy for requests to api_base;
  • (optional) prompt: Customize your prompt. This will appear in every chat request;

Console help (with tab-complete):

gptcli> .help -v

gptcli commands (use '.help -v' for verbose/'.help <topic>' for details):
======================================================================================================
.edit                 Run a text editor and optionally open a file with it
.help                 List available commands or provide detailed help for a specific command
.load                 Load conversation from Markdown/JSON file
.multiline            input multiple lines, end with ctrl-d(Linux/macOS) or ctrl-z(Windows). Cancel
                      with ctrl-c
.prompt               Load different prompts
.quit                 Exit this application
.reset                Reset session, i.e. clear chat history
.save                 Save current conversation to Markdown/JSON file
.set                  Set a settable parameter or show current settings of parameters
.usage                Tokens usage of current session / last N days, or print detail billing info

Run in Docker:

# build
$ docker build -t gptcli:latest .

# run
$ docker run -it --rm -v $PWD/.key:/gptcli/.key gptcli:latest -h

# for host proxy access:
$ docker run --rm -it -v $PWD/config.json:/gptcli/config.json --network host gptcli:latest -c /gptcli/config.json

Feature

  • Single Python script
  • Session based
  • Markdown support with code syntax highlight
  • Stream output support
  • Proxy support (HTTP/HTTPS/SOCKS4A/SOCKS5)
  • Multiline input support (via .multiline command)
  • Save and load session from file (Markdown/JSON) (via .save and .load command)
  • Print tokens usage in realtime, and tokens usage for last N days, and billing details
  • Integrate with llama_index to support chatting with documents

LINK