Datasets:

facebook

/

wiki_dpr

like 33

Follow

AI at Meta 3,337

I tag the people who appear in the dataset card as having added this dataset: https://huggingface.co/datasets/wiki_dpr#contributions
@thomwolf , @lewtun , @lhoestq

They all have the same text content (passages split every 100 words), but different embeddings and indexes.

@lhoestq but "no_embeddings" configs have no embeddings or indexes, only text content.

Their only features are "id", "text" and "title". See: https://huggingface.co/datasets/wiki_dpr/blob/b24a417d802a583f8922946c1c75210290e93108/wiki_dpr.py#L123-L125

Yes correct, those ones are the same indeed

@albertvillanova The configs listed above still load different precomputed index files, so they are not completely identical.

(datasets can still load an index even if the index name is not in the column list)