https://github.com/atroyn/math-llm

Grounding LLM mathematical reasoning with proof assistants.
https://github.com/atroyn/math-llm

Last synced: 16 days ago
JSON representation

Grounding LLM mathematical reasoning with proof assistants.

• Host: GitHub
• URL: https://github.com/atroyn/math-llm
• Owner: atroyn
• License: mit
• Created: 2023-06-09T23:16:39.000Z (over 1 year ago)
• Default Branch: main
• Last Pushed: 2023-06-16T05:18:54.000Z (over 1 year ago)
• Last Synced: 2024-05-23T04:37:59.452Z (6 months ago)
• Language: Python
• Homepage:
• Size: 5.86 KB
• Stars: 56
• Watchers: 2
• Forks: 4
• Open Issues: 3
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# Math LLM

This is a prototype project for grounding mathematical reasoning for large language models, using software proof assistants.

This repo is _highly_ experimental, and anything might break at any time or not work for you.

Contributions are welcome.

## Installation

You will need the `coqtop` command line application. This is a CLI for interacting with the [Coq](https://coq.inria.fr/download) proof assistant software. I used the default installation path and it worked out of the box following the instructions in the included README.

You will need an [OpenAI API Key](https://platform.openai.com/account/api-keys) set in the OPENAI_API_KEY environment variable.

Then run `pip install -r requirements.txt`

## Run

`python main.py`

## Roadmap

- [x] Interactively use Coq from Python

- [x] Use LLM to generate proofs in a readable format

- [x] Execute generated Coq

- [x] Use LLM to evaluate proofs for logical consistency

- [ ] Store proof results in memory

- [ ] Generate new conjectures

- [ ] Break down proofs recursively into sub-goals

- [ ] ...