| 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 |