cadnav / tooldefgenerator

Generates a structure suitable for the tools argument in a client.chat.completions.create call

ToolDefGenerator

ToolDefGenerator is a Python package designed to automatically generate tool descriptions suitable for the tools argument in a client.chat.completions.create call. It gathers the data by introspection.

Features

• Automatic Tool Description: Generates tool descriptions from Python function definitions
• Type Mapping: Maps Python data types to a predefined set of string representations.
• Function Name Mapping: Maps function names.
• Strict Mode: Enforces strict requirements for docstrings and type annotations.

Requirements

This package requires typing.Annotated to be available, so Python 3.9.0 and higher is required.

Installation

You can install ToolDefGenerator by cloning the repository and installing it via pip:

git clone https://github.com/yourrepository/ToolDefGenerator.git
cd ToolDefGenerator
pip install .

When working on the package it is useful to install it in editable form:

pip install -r requirements.txt
pip install -e .

Then test:

pytest -v tests

Usage

To use ToolDefGenerator, you need to instantiate it and pass it the functions you want to introspect to the generate method.

Example

from tool_def_generator import ToolDefGenerator
from typing import Annotated


# Define the functions
def example_function(param1: Annotated[str, "The first parameter"],
                     param2: Annotated[str, "The second parameter"]) -> str:
    """
    This is an example function.
    """
    pass


def example_function2(param1: Annotated[str, "The first parameter"],
                      param2: Annotated[str, "The second parameter"]) -> str:
    """
    This is another example function.
    """
    pass


# Create an instance of ToolDefGenerator
generator = ToolDefGenerator()

# Generate tool descriptions using the new generate method
tools = generator.generate(example_function, example_function2)

# Use the generated tools in your client.chat.completions.create call
response = client.chat.completions.create(
    model="gpt-3.5-turbo-1106",
    messages=messages,
    tools=tools,
    tool_choice="auto",
)

For more examples see tests.

API Reference

Class ToolDefGenerator

A class for generating descriptions of Python functions.

__init__(self, type_map=None, strict=True, name_mappings: List[Tuple[str, str]] = None)

Initializes the ToolDefGenerator.

• type_map: Maps Python types to strings (default provided).
• strict: Enforces strict checking for annotations and docstrings.
• name_mappings: List of tuples for custom name mappings of functions.

generate(self, *functions: Callable) -> list

Generates a description array for given functions.

• *functions: Functions to introspect.
• Returns: List of tool descriptions for each function.

Internal Method introspect(self, function: Callable)

Introspects a given function (used internally).

• function: The function to introspect.
• Returns: Dictionary with the function's name, description, and parameters.

TODO

• use typing-extensions package to get access to typing.Annotated in older versions of Python

Contributing

Contributions are welcome! Please feel free to submit pull requests, report bugs, or suggest features.

License

This project is licensed under the MIT License.