fietje-2-data / README.md
BramVanroy's picture
Update README.md
072edee verified
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.