OpenAI Functions

pip install funcmodels

from funcmodels import openai_function

@openai_function

Highlights

This documentation assumes you're already familiar with OpenAI function calling, and Pydantic BaseModel.

from typing import Literal

@openai_function
def get_stock_price(ticker: str, currency: Literal["USD", "EUR"] = "USD"):
    """
    Get the stock price of a company, by ticker symbol

    Parameters
    ----------
    ticker
        The ticker symbol of the company
    currency
        The currency to use
    """
    return f"182.41 {currency}, -0.48 (0.26%) today"


get_stock_price

OpenaiFunction({
    "name": "get_stock_price",
    "description": "Get the stock price of a company, by ticker symbol",
    "parameters": {
        "properties": {
            "ticker": {
                "type": "string",
                "description": "The ticker symbol of the company"
            },
            "currency": {
                "default": "USD",
                "enum": [
                    "USD",
                    "EUR"
                ],
                "type": "string",
                "description": "The currency to use"
            }
        },
        "required": [
            "ticker"
        ],
        "type": "object"
    }
})

@openai_function dynamically creates a custom pydantic.BaseModel class with:

• Your function's parameters as attributes, for validation
• Class attribute, schema, with an OpenAI Function object for your function
  • Parses docstring for description, and parameter descriptions, if present.
  • Type structure based on pydantic's .model_json_schema()
• Class method, .from_json() to easily instantiate your model from raw JSON arguments received from OpenAI
• A .__call__() method to easily call your original function, using the model's validated attributes.

Get our OpenAI function definition dictionary

get_stock_price.schema

{'name': 'get_stock_price', 'description': 'Get the stock price of a company, by ticker symbol', 'parameters': {'properties': {'ticker': {'type': 'string', 'description': 'The ticker symbol of the company'}, 'currency': {'default': 'USD', 'enum': ['USD', 'EUR'], 'type': 'string', 'description': 'The currency to use'}}, 'required': ['ticker'], 'type': 'object'}}

Instantiate our pydantic model, validating arguments

model = get_stock_price(ticker="AAPL")

Or, go directly from raw json arguments from OpenAI

raw_arguments_from_openai = '{"ticker": "AAPL"}'
model = get_stock_price.from_json(raw_arguments_from_openai)
model.currency

'USD'

Call our function, with already-validated arguments

model()

'182.41 USD, -0.48 (0.26%) today'

If you prefer Pydantic syntax, we can achieve the same thing using Fields

from pydantic import Field

@openai_function
def get_stock_price(
    ticker: str = Field(description="The ticker symbol of the company"),
    currency: Literal["USD", "EUR"] = Field("USD", description="The currency to use."),
):
    "Get the stock price of a company, by ticker symbol"
    return f"182.41 {currency}, -0.48 (0.26%) today"

Here, the field descriptions are defined in the parameters themselves, rather than the docstring.

The result is the exact same function definition as before:

get_stock_price

OpenaiFunction({
    "name": "get_stock_price",
    "description": "Get the stock price of a company, by ticker symbol",
    "parameters": {
        "properties": {
            "ticker": {
                "type": "string",
                "description": "The ticker symbol of the company"
            },
            "currency": {
                "default": "USD",
                "enum": [
                    "USD",
                    "EUR"
                ],
                "type": "string",
                "description": "The currency to use"
            }
        },
        "required": [
            "ticker"
        ],
        "type": "object"
    }
})