RevLLM

Reverse Engineering Tools for Large Language Models

Introduction

RevLLM is a Python library designed to facilitate the analysis of Transformer Language Models, particularly focusing on generative, decoder-type transformers. Our library aims to democratize the access to advanced explainability methods for data scientists and machine learning engineers who work with language models. Built on top of Andrej Karpathy's esteemed nanoGPT, RevLLM stands as a robust and user-friendly tool in the field of natural language processing.

Features

• Model Analysis: Deep insights into transformer language models.
• Explainability Tools: Easy-to-use methods to understand model decisions.
• Compatibility with nanoGPT: Built upon a reliable and efficient base.
• User-Friendly Interface: Designed for both novice and expert users.

Installation

To install RevLLM, simply run the following command in your Python environment:

./make_mamba_env.sh

License

RevLLM is released under the MIT License.

Acknowledgements

Special thanks to Andrej Karpathy and contributors to the nanoGPT project for their foundational work in the field.