OpenMathInstruct-2 (formatted) is an instruction-tuning corpus for mathematics containing 13,972,791 examples. The dataset, created by author mssfj, was last updated on December 12, 2025. Each entry includes a natural-language question and an answer with an explicit chain-of-thought reasoning section.
Use Cases
- Fine-tuning language models for math problem-solving based on the question-answer format.
- Training models to generate explicit reasoning steps based on the chain-of-thought traces.
- Benchmarking model performance on grade-school to advanced math topics mentioned in the description.
- Developing educational AI tutors using the algebra, geometry, and word problem content.
Strengths
- Contains 13,972,791 examples, providing substantial scale for training.
- Covers a range of math topics from grade-school to advanced reasoning, algebra, and geometry.
- Includes explicit chain-of-thought reasoning traces within the answer format.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
Provenance
- Source
- huggingface
- Collection Method
- Derived from OpenMathInstruct-2 and converted into a question-answer format.
- Time Range
- null
- Freshness
- Last updated 2025-12-12 21:47:19; freshness should be verified.
- Geography
- null