The Pile: An 800GB Dataset of Diverse Text for Language Modelingarxiv.org2 points·andyxor··1 commentOpen articleSaveView on HN