GPT-4 Function Python Decorator

This Python package creates a function decorator @openaifunc which can be used to automatically generate the functions parameter for the ChatGPT API.

The original code was generated with GPT-4. I'm new to Python and have never created a Python package.

Inspired by @memespdf on @sentdex YouTube-video comments

How to use

First, import the package at the top of your Python code:

from openai_decorator.openai_decorator import openaifunc, get_openai_funcs

Then, add a @openaifunc decorator to the functions you want to use with ChatGPT:

@openaifunc
def add_numbers(a: int, b: int):
    """
    This function adds two numbers.
    """
    return a + b

Then, you can get a list of all the functions and their definitions for ChatGPT with get_openai_funcs() like so:

response = openai.ChatCompletion.create(
    model="gpt-4-0613",
    messages=messages,
    functions=get_openai_funcs(),
    function_call="auto",
)

Parameter descriptions

As far as I know, there is no "official" way to add docstrings for parameters in Python, but you can add the parameter definitions to the docstring in PHP DocBlock style, and GPT-4 seems to obey them.

@openaifunc
def get_current_weather(location: str, country: str) -> str:
    """
    Gets the current weather information
    @param location: The location for which to get the weather
    @param country: The country in which to look for the location
    """

    if location is None:
        return "A location must be provided. Please ask the user which location they want the weather for"
    else:
        return "The weather is nice and sunny"

Currently, this will not populate the description of the parameters in the API request, but GPT-4 still adheres to the rules.

Pydantic Models

You can also set descriptions for the function parameters with Pydantic models. This will actually populate the description of the parameters in the API request.

from pydantic import BaseModel, Field

class LocationModel(BaseModel):
    location: str = Field(
        description="The location for which to get the weather"
    )
    country: str = Field(
        description="The country in which to look for the location"
    )

@openaifunc
def get_current_weather(location: LocationModel) -> str:
    """
    Gets the current weather information
    """
    location = LocationModel.parse_obj(location)

    if location is None:
        return "A location must be provided. Please ask the user which location they want the weather for"
    else:
        return "The weather is nice and sunny"

Chatbot

There's a demo chatbot that uses the GPT-4 API with function calling. You can run it by exporting your OpenAI API key first:

$ export OPENAI_API_KEY=YOUR_API_KEY

And then running the script:

$ ./chatbot.py

You can test it by asking it about the weather, some YouTube channel recommendations or to calculate the length of a string.

You can also modify the functions in the chatbot code, to test your own functions.