metadata
dataset_info:
features:
- name: input_ids
sequence: int32
splits:
- name: train
num_bytes: 5881094452
num_examples: 5720909
- name: test
num_bytes: 309531828
num_examples: 301101
download_size: 3050952379
dataset_size: 6190626280
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: test
path: data/test-*
The first 15% of Openwebtext2 tokenized with LLamaTokenizer for 256 context length. Total of ~1.5 Billion tokens.