Pantograph/README.md

PyPantograph

A Machine-to-Machine Interaction System for Lean 4.

Installation

1. Install poetry
2. Clone this repository with submodules:

git clone --recursive-submodules <repo-path>

3. Install elan and lake: See Lean Manual
4. Execute

poetry build
poetry install

Documentation

Build the documentations by

jupyter-book build docs

Then serve

cd docs/_build/html
python3 -m http.server -d .

Examples

For API interaction examples, see examples/README.md

An agent based on the sglang library is provided in pantograph/search_llm.py. To use this agent, set the environment variable OPENAI_API_KEY, and run

python3 -m pantograph.search_llm

Experiments

In experiments/, there are some experiments:

1. minif2f/ is an example of executing a SGLANG based prover on the miniF2F dataset
2. dsp is an Lean implementation of Draft-Sketch-Prove

If the experiments don't work, run them in poetry shell. The environment variable OPENAI_API_KEY must be set when running experiments calling the OpenAI API.

Referencing

@misc{pantograph,
	title = "Pantograph, A Machine-to-Machine Interface for Lean 4",
	author = {Aniva, Leni and Miranda, Brando and Sun, Chuyue},
	year = 2024,
	howpublished = {\url{https://github.com/lenianiva/PyPantograph}}
}