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