metadata
dataset_info:
features:
- name: input_ids
sequence: int32
- name: attention_mask
sequence: int8
- name: labels
sequence: int64
splits:
- name: train
num_bytes: 366738831440
num_examples: 13768540
- name: test
num_bytes: 479847540
num_examples: 18015
download_size: 105689403490
dataset_size: 367218678980
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: test
path: data/test-*
The data to train fietje-2. It contains around 28B tokens taken from the Dutch parts of wikipedia and CulturaX.