The Pile: An 800GB Dataset of Diverse Text for Language Modeling

GPTKB entity