Datasets:

allenai

/

c4

like 170

Finally took the time to configure the dataset to be loaded in 🤗 Datasets.
This will also add a dropdowns to the Dataset Viewer with "en", "en.noclean", "en.noblocklist", "realsnewslike", "multilingual", as well as each mC4 language "af", "ar", "ar", etc.

>>> from datasets import load_dataset
>>> load_dataset("allenai/c4", "en", revision="refs/pr/6", streaming=True)
IterableDatasetDict({
    train: IterableDataset({
        features: ['text', 'timestamp', 'url'],
        n_shards: 1024
    })
    validation: IterableDataset({
        features: ['text', 'timestamp', 'url'],
        n_shards: 8
    })
})

Bonus: I documented how to load it not only with 🤗 Datasets but also with Dask.

Ultimately I think we can redirect the c4 and mc4 pages to this one