Datasets:

wikimedia

/

wikipedia

like 382

yes, the size is smaller because of preprocessing

I have a similar question since I noticed that it is written in the README that it is a train "subset" for each language. Is it just some sampling of wikipedia pages or all should be there? I would need all the pages on medical subjects. I am planning to filter the rest out for some application.

The dataset contains all the Wikipedia pages at the date of the dump.

Each date-language subset contains a single "train" split.

Although it is worth noting that the default preprocessing script removes all the pages that redirect to another page, among other things. @TiamoLee