This repo implements the ProofAug method introduced in our work "ProofAug: Efficient Neural Theorem Proving via Fine-grained Proof Structure Analysis".

The Isabelle implementation of ProofAug achieves a pass rate of 66.0% on miniF2F-test in 2100 attempts using the deepseek-math-7b-base model, setting a new state of the art among existing Isabelle-based methods. Also, the Lean version of ProofAug that can improve the pass@1 performance of Kimina-Prover-Preview-Distill-1.5B by 6.1%.

See isabelle_src for the Isabelle implementation and lean_src for the Lean implementation.

If you have any questions, you can raise an issue or contact [email protected].

If you find this work useful , please consider citing:

@article{liu2025proofaug,
  title={ProofAug: Efficient Neural Theorem Proving via Fine-grained Proof Structure Analysis},
  author={Liu, Haoxiong and Sun, Jiacheng and Li, Zhenguo and Yao, Andrew C},
  journal={arXiv preprint arXiv:2501.18310},
  year={2025}
}