README.md

metadata

dataset_info:
  features:
    - name: text
      dtype: string
    - name: id
      dtype: string
    - name: token_count
      dtype: int64
    - name: corrupt
      dtype: string
  splits:
    - name: train
      num_bytes: 3812219177.6666665
      num_examples: 385024
    - name: test
      num_bytes: 81111046.33333333
      num_examples: 8192
  download_size: 3223676991
  dataset_size: 3893330224
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*

Quest-corruption 393K (CLM)

This is quest-corruption for training of CLM restorers. For RM training, use the data in quest-corruption-1m-rm.