CallingGPT

GPT's Function Calling Demo, a experiment of self-hosted ChatGPT-Plugins-like platform.

Recommend reading: function-calling

Abstract

OpenAI's GPT models provide a function calling feature, so we can easily create ChatGPT-Plugins-like tools. This repository is a proof-of-concept of the function calling feature.
In this experiment, we defined the Plugin as Namespace which contains a serial of functions. While user performing a conversation, the functions in Namespace will be called by the API and return the result to the user.

Usage

1. Clone this repository and install the dependencies.

   git clone https://github.com/RockChinQ/CallingGPT
   cd CallingGPT
   pip install -r requirements.txt
   

2. Run the main.py to generate config.yaml

   python main.py
   

3. Edit the config.yaml to set your API key and other settings.

4. Run the main.py and pass your modules.

   python main.py <module0> <module1> ...
   

Example

Use the example/greet.py, provides a greet function called when user ask GPT to greet someone.

python main.py example/greet.py

Then you can talk to the bot.

$ python main.py examples/greet.py 
Using module: examples.greet
>>> Hello and who are you?
<<< Hello! I am an AI assistant. How can I assist you today?
>>> say hello to Rock
call<examples-greet-greet>: {
  "user": "Rock"
}
<<< Hello, Rock! How can I assist you today?
>>> and to Alice
call<examples-greet-greet>: {
  "user": "Alice"
}
<<< Hello, Alice! How can I assist you today?
>>>

Type help to get help.
See wiki for the function format.

Other Examples

python main.py examples/draw_and_wrapper_md.py 

$ python main.py examples/draw_and_wrapper_md.py 
Using module: examples.draw_and_wrapper_md
>>> hello!
<<< Hi there! How can I assist you today?
>>> draw a sunset for me please
call<examples-draw_and_wrapper_md-draw>: {
  "prompt": "sunset"
}
<<< Sure! Here's a beautiful sunset for you:

![Sunset](https://oaidalleapiprodscus.blob.core.windows.net/private/org-VS9HEpJba78GXVfOcmVo7qaM/user-OHa7Jo3kL4XJDg9lo7AzdWNT/img-QmDUiwp1IGFcu8pDGZh0i7r8.png)

I hope you like it! Let me know if there's anything else I can help you with.
>>> 

For Code

1. Install the package

   pip install --upgrade CallingGPT
   

2. Create your own functions in modules(these modules can also be used in the CLI mode)

   # your_module_a.py
   def func_a(prompt: str) -> str:  # Type hint of EACH argument and return value is REQUIRED.
       """
       The description of this func a, will be provided to the api.
   
       Args:
           prompt(str): The prompt of the function.
   
       Returns:
           The result of the function.
       """
       # Google style docstring is REQUIRED, it will be split into
       # `description` and `params`(required if there are args) and 
       # `returns`(optional), `\n\n` between each part.
       return "func_a: " + prompt
   

   # your_module_b.py
   def adder(a: int, b: int) -> int:
       """
       Add two numbers.
   
       Args:
           a: The first number.
           b: The second number.
   
       Returns:
           The sum of a and b.
       """
       # Type hints of args in docstring is optional.
       return a + b
   

3. Call the wrapper

   from CallingGPT.session.session import Session
   import your_module_a, your_module_b
   import openai
   
   openai.api_key = 'your_openai_api_key'
   
   session = Session([your_module_a, your_module_b])
   
   for reply in session.ask("your prompt"):
       # session.ask will yield each time the api returns a result,
       # before calling function, it will print the function name and args.
       # e.g. here's a function call:
       # {
       #   "role": "assistant",
       #   "content": null,
       #   "function_call": {
       #     "name": "examples-draw_and_wrapper_md-draw",
       #     "arguments": "{\n  \"prompt\": \"cat\"\n}"
       #   }
       # }
       # 
       # while here's a normal reply:
       # {
       #   "role": "assistant",
       #   "content": "Hello, I am an AI assistant. How can I assist you today?"
       # }
       print(reply)
   

   Session will automatically manage context for you.

See wiki for the function format.