A arxiv.org The Pile: An 800GB dataset of diverse text for language modeling (2020) 184 points charlysl 3 years ago 70 comments
