metadata

configs:
  - config_name: strict
    default: true
    data_files:
      - split: train
        path:
          - train_100M/bnc_spoken.csv
          - train_100M/childes.csv
          - train_100M/gutenberg.csv
          - train_100M/open_subtitles.csv
          - train_100M/simple_wiki.csv
          - train_100M/switchboard.csv
      - split: valid
        path:
          - dev/bnc_spoken.csv
          - dev/childes.csv
          - dev/gutenberg.csv
          - dev/open_subtitles.csv
          - dev/simple_wiki.csv
          - dev/switchboard.csv
  - config_name: strict_small
    data_files:
      - split: train
        path:
          - train_10M/bnc_spoken.csv
          - train_10M/childes.csv
          - train_10M/gutenberg.csv
          - train_10M/open_subtitles.csv
          - train_10M/simple_wiki.csv
          - train_10M/switchboard.csv
      - split: valid
        path:
          - dev/bnc_spoken.csv
          - dev/childes.csv
          - dev/gutenberg.csv
          - dev/open_subtitles.csv
          - dev/simple_wiki.csv
          - dev/switchboard.csv
language:
  - en
tags:
  - language modeling
  - cognitive modeling
pretty_name: BabyLM Phonemized
size_categories:
  - 1M<n<10M

Phonemized BabyLM Pre-training Data

This corpus contains the strict portion of the BabyLM 2024 pre-training data converted to phonemes using Corpus Phonemizer. The original orthographic data is also available.