FineProofs SFT: 7,777 Olympiad Math Problems with Formal Proofs

FineProofs SFT contains 7,777 mathematical Olympiad samples featuring chain-of-thought reasoning and formal proofs distilled from DeepSeek-Math-V2. Developed by lm-provers and updated in February 2026, the dataset sources 4,300 unique problems from international competitions and Art of Problem Solving (AoPS).

Use Cases

• Fine-tuning language models for mathematical reasoning using the provided chain-of-thought traces
• Training automated theorem provers to generate formal proofs for competition-level problems
• Evaluating the ability of LLMs to solve complex problems from international Olympiad competitions

Strengths

• Contains 7,777 samples covering 4,300 unique high-level Olympiad problems
• Includes formal proofs distilled from the DeepSeek-Math-V2 model
• Sourced from reputable competition archives and Art of Problem Solving (AoPS)

Limitations

• Reasoning traces are synthetic (distilled from an LLM) rather than human-authored
• Relatively small sample size for broad mathematical domain coverage

Provenance

Source

International Olympiad competitions and Art of Problem Solving (AoPS)

Collection Method

Distilled from DeepSeek-Math-V2 and curated from competition archives

Freshness

Last updated February 2026.

Geography

Global (International Olympiads)