Use the following command to load this dataset in TFDS:

ds = tfds.load('huggingface:gigaword')
  • Description:
Headline-generation on a corpus of article pairs from Gigaword consisting of
around 4 million articles. Use the 'org_data' provided by which is identical to but with better format.

There are two features:
  - document: article.
  - summary: headline.
  • License: No known license
  • Version: 1.2.0
  • Splits:
Split Examples
'test' 1951
'train' 3803957
'validation' 189651
  • Features:
    "document": {
        "dtype": "string",
        "id": null,
        "_type": "Value"
    "summary": {
        "dtype": "string",
        "id": null,
        "_type": "Value"