chunked-fineweb-edu / README.md
gabrielagc's picture
Upload dataset
5a09352 verified
metadata
dataset_info:
  features:
    - name: text
      dtype: string
    - name: id
      dtype: string
    - name: token_count
      dtype: int64
    - name: preceding_token_count
      dtype: int64
  splits:
    - name: train
      num_bytes: 5862530740.527941
      num_examples: 2658953
  download_size: 3393249830
  dataset_size: 5862530740.527941
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

Samples from the FineWeb-Edu dataset divided into chunks of 512 tokens. The beginning of each sequence (first 20 tokens) is prepended to the subsequent chunks. Only samples with a token count greater than 80% of 512 have been included in the final dataset.