baby-python / README.md
squidWorm's picture
updated babylm dataset link
55ac20a verified
|
raw
history blame
1.34 kB
metadata
language:
  - en
license: mit
task_categories:
  - text-generation
pretty_name: Baby Python
tags:
  - python
  - english
dataset_info:
  features:
    - name: content
      dtype: string
    - name: type
      dtype: string
  splits:
    - name: train
      num_bytes: 1215922226
      num_examples: 10695758
    - name: validation
      num_bytes: 117555692
      num_examples: 1037288
  download_size: 605995775
  dataset_size: 1333477918
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*

Baby Python

English + Python pretraining dataset.

This dataset contains the dataset from the BabyLM 100M dataset, evenly matched by a truncated subset of the small-python-stack.

English / all of BabyLM 100M

  • CHILDES (child-directed speech)
  • Subtitles (speech)
  • BNC (speech)
  • TED talks (speech)
  • children's books (simple written language)

Python