---
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](https://huggingface.co/BramVanroy/fietje-2). It contains around 28B tokens taken from the Dutch parts of [wikipedia](https://huggingface.co/datasets/wikimedia/wikipedia) and [CulturaX](https://huggingface.co/datasets/uonlp/CulturaX).