metadata
language:
- en
dataset_info:
features:
- name: provenance
dtype: string
- name: text
dtype: string
splits:
- name: train
num_bytes: 34959777487
num_examples: 13095416
download_size: 20602021674
dataset_size: 34959777487
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
Taken directly from allenai/dolma. Roughly 10B tokens. Useful for small scale testing or maybe continued pretraining.