README.md

metadata

language:
  - en
license: mit
task_categories:
  - text-generation
dataset_info:
  features:
    - name: pageid
      dtype: int64
    - name: title
      dtype: string
    - name: revid
      dtype: int64
    - name: description
      dtype: string
    - name: categories
      sequence: string
    - name: markdown
      dtype: string
    - name: sections
      sequence: string
    - name: token
      dtype: int64
    - name: __index_level_0__
      dtype: int64
  splits:
    - name: train
      num_bytes: 159522657
      num_examples: 18241
    - name: validate
      num_bytes: 28217967
      num_examples: 3220
    - name: test
      num_bytes: 10035346
      num_examples: 1130
  download_size: 114465310
  dataset_size: 197775970
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validate
        path: data/validate-*
      - split: test
        path: data/test-*

• modified version of Goodwiki Dataset from https://huggingface.co/datasets/euirim/goodwiki
• added section column (markdown headings), token column (computed with tiktoken)
• only contains articles with less than 3,000 token
• train, validate, test split (80, 15, 5)