Nemotron Math Proofs V2: 82,737 Mathematical Proof and Verification Traces

Nemotron-Math-Proofs-v2 contains 82,737 samples of mathematical proof-generation, verification, and meta-verification traces. The problems are sourced from the Art of Problem Solving subset of the nvidia/Nemotron-Math-Proofs-v1 dataset. Proofs were generated using DeepSeek-V4-Pro on Max inference mode.

Use Cases

• Training models for mathematical proof generation based on the described proof traces.
• Benchmarking automated proof verification systems using the verification traces.
• Studying meta-verification techniques for mathematical reasoning in AI systems.
• Fine-tuning language models on structured mathematical problem-solving from the AoPS domain.

Strengths

• Contains 82,737 samples across 5,752 unique problems, providing substantial volume.
• Focuses on a curated subset from the Art of Problem Solving, a known source for challenging math problems.
• Includes multiple trace types: proofs, verification traces, and meta-verification traces.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

nvidia/Nemotron-Math-Proofs-v1 (AoPS subset)

Collection Method

Proofs generated using DeepSeek-V4-Pro on Max inference mode.

Freshness

Last updated 2026-06-04 04:46:34; freshness should be verified.