mxab / taifun

A simple framework for creating typed OpenAI functions from Python functions

Taifun - Typed AI Functions

A simple framework for creating typed AI functions. It inspects the function's docstring and parameters to provide functions for OpenAI's API. You can either use the Taifun class to create a dict of functions to pass to OpenAI's API or use the TaifunConversationRunner class to run a conversation with functions.

Installation

pip install taifun

Usage

Initialize a Taifun instance and decorate your functions with @taifun.fn()

taifun = Taifun()

@taifun.fn()
def weather_forcast(location: str) -> str:
    """
    Get the weather forcast for a given location

    Parameters
    ----------
    location: str
        the user's location like a Ciry and State, e.g. San Francisco, CA

    """

    return f"The weather in {location} is rainy"

Then you can use the functions_as_dict method to get a dict that can be passed to OpenAI's functions field

functions = taifun.functions_as_dict()

Then you can use the handle_function_call method to handle a function call from OpenAI's API

functions is a dict that can be passed to OpenAI's functions field

[
 {
  "name": "weather_forcast",
  "description": "Get the weather forcast for a given location",
  "parameters": {
   "properties": {
    "location": {
     "description": "the user's location like a Ciry and State, e.g. San Francisco, CA",
     "title": "Location",
     "type": "string"
    }
   },
   "required": [
    "location"
   ],
   "title": "FunctionParameters",
   "type": "object"
  }
 }
]

Pass functions to the functions field of OpenAI's API

messages = [...]
result = openai.ChatCompletion.create(
    model="gpt-4",
    messages=messages,
    functions=functions,
    function_call="auto",
)

If the response from OpenAI's API has a function call, you can handle it with handle_function_call

function_call = result["choices"][0]["message"].get("function_call")

if function_call is not None:
    # handle the function call
    function_response = taifun.handle_function_call(function_call)
    # return the function response
    messages.append(
        {
            "role": "function",
            "name": function_call["name"],
            "content": function_response,
        }
    )
    # reply

Demo with functions to pass to OpenAI

taifun = Taifun()


@taifun.fn()
def weather_forcast(location: str) -> str:
    """
    Get the weather forcast for a given location

    Parameters
    ----------
    location: str
        the location to get the weather forcast for

    """

    # random weather
    weather = random.choice(["sunny", "rainy", "cloudy", "snowy"])

    return f"The weather in {location} is {weather}"


messages = [
    {
        "role": "user",
        "content": "Is it rainingy in berlin today?",
    },
]

# export functions as json schema dict for openai
functions = taifun.functions_as_dict()


result = openai.ChatCompletion.create(
    model="gpt-4",
    messages=messages,
    functions=functions,
    function_call="auto",
)
response_message = result["choices"][0]["message"]

print(f"assistant: {response_message['content']}")

function_call = response_message.get("function_call")

messages.append(response_message)
if function_call is not None:
    # handle the function call
    function_response = taifun.handle_function_call(function_call)

    # responed with the function response
    print(f"function response: {function_response}")
    messages.append(
        {
            "role": "function",
            "name": function_call["name"],
            "content": function_response,
        }
    )

    result2 = openai.ChatCompletion.create(
        model="gpt-4",
        messages=messages,
        functions=functions,
        function_call="auto",
    )
    response_message2 = result2["choices"][0]["message"]
    print(f"assistant: {response_message2['content']}")

A full example including the TaifunConversationRunner

taifun = Taifun()


@taifun.fn()
def get_location() -> str:
    """
    Get the user's location

    returns: the user's location like a Ciry and State, e.g. San Francisco, CA
    """
    location = Prompt.ask("What is your location?")

    return location


@taifun.fn()
def get_lang_lat(location: str) -> dict:
    """
    Get the latitude and longitude of a location

    Parameters
    ----------
    location: str 
        the user's location like a Ciry and State, e.g. San Francisco, CA

    """

    response = httpx.get(
        f"https://nominatim.openstreetmap.org/search/{urlparse.quote(location)}",
        params={
            "format": "json",
        },
    )
    response.raise_for_status()
    data = response.json()
    lat = data[0]["lat"]
    lng = data[0]["lon"]

    return {"latitute": lat, "longitude": lng}


class Coordinates(BaseModel):
    latitude: float = Field(
        ..., title="Latitude", description="The latitude of a location"
    )
    longitude: float = Field(
        ..., title="Longitude", description="The longitude of a location"
    )


@taifun.fn()
def get_current_weather(coordinates: Coordinates):
    """Get the current weather in a given longitude and latitude

    Parameters
    ----------
    coordinates: Coordinates
        the latitude and longitude of a location

    Returns:
        dict: a dictionary of the current weather

    """

    response = httpx.get(
        "https://api.open-meteo.com/v1/forecast",
        params={
            "latitude": coordinates.latitude,
            "longitude": coordinates.longitude,
            "current_weather": True,
        },
    )
    response.raise_for_status()
    data = response.json()
    return data


if __name__ == "__main__":
    openai.api_key_path = os.path.expanduser("~") + "/.openai_api_key"
    runner = TaifunConversationRunner(taifun)
    result = runner.run("Will I need an umbrella today?")

    rich.print(result)