HunyuanProver: A Scalable Data Synthesis Framework and Guided Tree Search for Automated Theorem Proving

TL;DR

HunyuanProver is a scalable, guided theorem prover based on a fine-tuned language model that achieves state-of-the-art results on major benchmarks and will release a large dataset for community use.

Contribution

It introduces a scalable data synthesis framework and guided tree search algorithms for automated theorem proving with a language model.

Findings

Achieves 68.4% pass rate on miniF2F-test, surpassing previous SOTA of 65.9%.

Successfully proves 4 IMO problems in the miniF2F benchmark.

Plans to release a dataset of 30,000 synthesized instances for community research.

Abstract

We introduce HunyuanProver, an language model finetuned from the Hunyuan 7B for interactive automatic theorem proving with LEAN4. To alleviate the data sparsity issue, we design a scalable framework to iterative synthesize data with low cost. Besides, guided tree search algorithms are designed to enable effective ``system 2 thinking`` of the prover. HunyuanProver achieves state-of-the-art (SOTA) performances on major benchmarks. Specifically, it achieves a pass of 68.4% on the miniF2F-test compared to 65.9%, the current SOTA results. It proves 4 IMO statements (imo_1960_p2, imo_1962_p2}, imo_1964_p2 and imo_1983_p6) in miniF2F-test. To benefit the community, we will open-source a dataset of 30k synthesized instances, where each instance contains the original question in natural language, the converted statement by autoformalization, and the proof by HunyuanProver.