metadata

configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
dataset_info:
  features:
    - name: input_ids
      sequence: int32
    - name: attention_mask
      sequence: int8
    - name: labels
      sequence: int64
  splits:
    - name: train
      num_bytes: 67744337720
      num_examples: 79514
  download_size: 1125510240
  dataset_size: 67744337720

Dataset Card for "pg_books-tokenized-bos-eos-chunked-65536"

The pg19 dataset tokenized under LLaMA into 64k chunks, bookended with BOS and EOS