OpenWebMath: An Open Dataset of High-Quality Mathematical Web Text

Abstract:

There is growing evidence that pretraining on high quality, carefully thought-out tokens such as code or mathematics plays an important role in improving the reasoning abilities of large language models. For example, Minerva, a PaLM model finetuned on billions of tokens of mathematical documents from arXiv and the web, reported dramatically improved performance on problems that require quantitative reasoning. However, because all known open source web datasets employ preprocessing that does not faithfully preserve mathematical notation, the benefits of …

abstract arxiv code dataset documents evidence example language language models large language large language models mathematics minerva palm quality reasoning role text thought tokens web