This Dataset is in ready to use format shuffled, contains 50B+ gpt2 tokens.
Designed for training Rubin-series models.
English-only dataset
Contains eval dataset (around 100 million tokens) and 50B training tokens.
Both are well shuffled into sequences of 1024 tokens.
The file is cleaned and contain raw tokens in unit16 format