open-web-math

Keiran Paster , Marco Dos Santos , Zhangir Azerbayev, Jimmy Ba GitHub ArXiv PDF OpenWebMath is a dataset containing the majority of the high-quality, mathematical text from the internet. It is filtered and extracted from over 200B HTML files on Common Crawl down to a set of 6.3 million documents containing a total of 14.7B tokens. OpenWebMath is intended for use in pretraining and finetuninglarge language models. You can download the dataset using Hugging Face: from datasets import… See the full description on the dataset page:

Source: Hugging Face Hub (open-web-math/open-web-math). Metadata imported from the dataset’s Hub tags.