Datasets:

wikipedia

like 485

Thanks.

I think we should also host the data files currently in GCS for reproducibility: users may want to reproduce results obtained by using those data files.

@albertvillanova Addressed, so feel free to review this PR again.

A few comments:

Dataset "wikipedia" is deprecated and will soon be deleted. Use the wikimedia/wikipedia dataset instead.

Maybe "will soon redirect to wikimedia/wikipedia" ?

Alternatively we can redirect it to wikimedia/wikipedia_script so avoid any breaking change - especially given the high number of downloads for this one. WDYT ?
Maybe it's also more visible than the "script" branch on wikimedia/wikipedia @albertvillanova

Use datasets.load_dataset('wikimedia/wikipedia', '20231101.en') to load the dataset instead.

Maybe "Use datasets.load_dataset('wikimedia/wikipedia', f'{date}.{lang_code}') to load the dataset instead (e.g. '20231101.en')" ?

Also what about suggesting to pass num_proc=to load_dataset in the dataset card for users that want to speed it up ?

Also what about suggesting to pass num_proc=to load_dataset in the dataset card for users that want to speed it up ?

Good point! I added a tip.

Also, as discussed offline, I removed the deprecation messages since this can be addressed later in a separate PR.