вики40б

ds = tfds.load('huggingface:wiki40b/en')

Clean-up text for 40+ Wikipedia languages editions of pages
correspond to entities. The datasets have train/dev/test splits per language.
The dataset is cleaned up by page filtering to remove disambiguation pages,
redirect pages, deleted pages, and non-entity pages. Each example contains the
wikidata id of the entity, and the full Wikipedia article after page processing
that removes non-content sections and structured objects.

{
    "wikidata_id": {
        "dtype": "string",
        "id": null,
        "_type": "Value"
    },
    "text": {
        "dtype": "string",
        "id": null,
        "_type": "Value"
    },
    "version_id": {
        "dtype": "string",
        "id": null,
        "_type": "Value"
    }
}