metadata

dataset_info:
  features:
    - name: article
      dtype: string
    - name: summary
      dtype: string
    - name: article_len_approx
      dtype: int64
    - name: summary_len_approx
      dtype: int64
    - name: orig_idx
      dtype: int64
  splits:
    - name: train
      num_bytes: 992678294
      num_examples: 3783821
    - name: validation
      num_bytes: 49616640
      num_examples: 188811
    - name: test
      num_bytes: 468746
      num_examples: 1822
  download_size: 567157084
  dataset_size: 1042763680
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
      - split: test
        path: data/test-*
task_categories:
  - summarization
language:
  - en
pretty_name: Gigatrue
size_categories:
  - 1M<n<10M

Gigatrue abstractive summarisation dataset.

It is a cleaned version of https://huggingface.co/datasets/Harvard/gigaword.
Added generated number values.
Applied truecasing using https://github.com/daltonfury42/truecase.