StackMathQA

The development of sophisticated mathematical reasoning in large language models is often limited by the scarcity of large-scale, high-quality, domain-specific training data. StackMathQA addresses this gap with nearly two million question-and-answer pairs from the Stack Exchange network, aggregating expert and enthusiast discussions from Mathematics Stack Exchange, MathOverflow, Cross Validated, and Physics Stack Exchange.

The dataset is available in multiple formats and curated subsets created through importance resampling, supporting use cases from large-scale continual pre-training to targeted supervised fine-tuning.

Read the paper🤗 Dataset

Formats and curated subsets

Data is released as one-question-to-many-answers (qalist) and flattened one-question-to-one-answer (1q1a) pairs. Importance-resampled subsets are available at 100K, 200K, 400K, 800K, and 1.6M examples, prioritizing higher-engagement material at every scale.

If you use StackMathQA in your research, please cite the technical report:

@techreport{zhang2024stackmathqa,
  title       = {StackMathQA: A Curated Collection of 2 Million Mathematical
                 Questions and Answers Sourced from Stack Exchange},
  author      = {Zhang, Yifan},
  year        = {2024},
  institution = {Math-AI},
  url         = {https://stackmathqa.github.io/StackMathQA.pdf}
}

Copy BibTeX