LeanDojo: Machine Learning for Theorem Proving in Lean

LeanDojo is a Python library for learning–based theorem provers in Lean, supporting both Lean 3 and Lean 4. It provides two main features:

• Extracting data (proof states, tactics, premises, etc.) from Lean repos.
• Interacting with Lean programmatically.

---

Requirements

• Supported platforms: Linux, Windows (WSL), and macOS (:warning: experimental for Apple silicon)
• Git >= 2.25
• 3.9 <= Python < 3.11
• wget
• Docker strongly recommended

Installation

LeanDojo is available on PyPI and can be installed via pip:

pip install lean-dojo

It can also be installed locally from the Git repo:

pip install .

Documentation

• Getting Started
• Demos: Lean 3, Lean 4
• Full documentation

Questions and Bugs

• For general questions and discussions, please use GitHub Discussions.
• To report a potential bug, please open an issue. In the issue, please include your OS information, the version of LeanDojo, the exact steps to reproduce the error, and complete logs in debug mode (setting the environment variable VERBOSE to 1). The more details you provide, the better we will be able to help you.

Related Links

• LeanDojo Website: The official website of LeanDojo.
• LeanDojo Benchmark : The dataset used in our paper, consisting of 98,641 theorems and proofs extracted from mathlib by generate-benchmark-lean3.ipynb.
• LeanDojo Benchmark 4 : The Lean 4 version of LeanDojo Benchmark, consisting of 100,780 theorems and proofs extracted from mathlib4 by generate-benchmark-lean4.ipynb.
• ReProver: The ReProver (Retrieval-Augmented Prover) model in our paper.
• LeanDojo ChatGPT Plugin
• LeanInfer: Running Neural Networks Directly in Lean

Citation

LeanDojo: Theorem Proving with Retrieval-Augmented Language Models
Under review, NeurIPS (Datasets and Benchmarks Track), 2023
Kaiyu Yang, Aidan Swope, Alex Gu, Rahul Chalamala,
Peiyang Song, Shixing Yu, Saad Godil, Ryan Prenger, Anima Anandkumar

@article{yang2023leandojo,
  title={{LeanDojo}: Theorem Proving with Retrieval-Augmented Language Models},
  author={Yang, Kaiyu and Swope, Aidan and Gu, Alex and Chalamala, Rahul and Song, Peiyang and Yu, Shixing and Godil, Saad and Prenger, Ryan and Anandkumar, Anima},
  journal={arXiv preprint arXiv:2306.15626},
  year={2023}
}