Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

1 Commit
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Testing the efficacy of LLMs in writing Lean4

This repo contains the data and the scripts used for the paper.

Scripts

In this repo the script main.py is provided. its purpose is to filter lean4_workbook theorems and retrieve them, called without args it returns a list of proven theorems with the associated index (referenced in the data). Called with a theorem index it will return all the details about that specific theorem.

Data

Theorems and proofs on which LLMs are tested are in the lean4 workbook, that can by retrieved by git cloning it:

git clone https://huggingface.co/datasets/internlm/Lean-Workbook

Results of our testing are contained in data.csv and summarized in data_summary.csv. livebench.csv contains the current (12/11/2024) benchmarks results for the tested LLMs.

Prompts

The following prompts have been used to test the LLMs.

Translation

translate this statement in lean4, don't prove it and provide only lean4 translation without description

<lean_workbook natural language statement>

Proving

proof this in lean4, explain the code in depth step by step and give the complete solution as code at the end

<lean_workbook lean4 statement>

Understanding

is this a valid Lean4 proof?

<lean_workbook statement and proof in lean4>

About

No description, website, or topics provided.

Resources

Stars

2 stars

Watchers

1 watching

Forks

Releases

Packages

Contributors

Languages