GRAD: 1,933 Graduate-Level Synthetic Math Problems with Proofs

GRAD is a high-quality synthetic mathematics dataset containing 1,933 original problems at graduate and research level, each accompanied by a complete, detailed, step-by-step proof in clear mathematical English and LaTeX. All problems and proofs were generated from scratch by Xerv-AI and released in December 2025. No entry has ever appeared in textbooks, competition archives, or research papers.

Use Cases

• Fine-tuning language models for advanced mathematical reasoning based on the graduate-level problem statements.
• Benchmarking automated theorem provers based on the dataset's detailed step-by-step proofs.
• Training models for synthetic data generation in mathematics based on the described original, non-derivative content.
• Studying formal language representation in LaTeX based on the proof formatting mentioned in the description.

Strengths

• Contains 1,933 original, high-quality problems generated from scratch.
• Each problem includes a complete, detailed, step-by-step proof in mathematical English and LaTeX.
• Problems are at graduate and research level, targeting advanced reasoning.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Data may reflect bias inherent to its synthetic generation process.

Provenance

Source

Xerv-AI

Collection Method

Synthetically generated from scratch.

Freshness

Last updated 2025-12-14 03:15:59; freshness should be verified.