Brokenarxiv Training: Research-Level Mathematical Problems from ArXiv

Training data generated from past ArXiv articles using the BrokenArXiv pipeline. The dataset is intended for training models on research-level mathematical problems and is licensed under cc-by-4.0, though individual rows may have different licenses. The dataset was created by MathArena and was last updated on 2026-06-16.

Use Cases

• Train language models on mathematical problem-solving based on research-level content.
• Fine-tune models for parsing and understanding academic mathematics from ArXiv articles.
• Benchmark model performance on complex, domain-specific text generation tasks.

Strengths

• Derived from the established ArXiv preprint repository, suggesting a source of academic content.
• Explicitly intended for training models on research-level mathematical problems.
• License information is provided, with a base cc-by-4.0 license noted.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Freshness should be verified as the last update timestamp is 2026-06-16.

Provenance

Source

ArXiv articles processed via the BrokenArXiv pipeline by MathArena.

Collection Method

Generated from past ArXiv articles.

Freshness

Last updated 2026-06-16 14:12:54.