Add OSCAR dataset card

[pjox]

I added more information and completed the dataset card for OSCAR which was started by @lhoestq in his previous PR.

[lhoestq]

Awesome thank you :)

I added a few suggestions.

I'm also generating the tags that go at the top of the dataset card

[pjox]

@lhoestq Thanks for the suggestions! I agree with all of them. Should I accept them one by one or can I accept them all at once? When I try to load the whole diff GitHub is complaining and it does no render them well (probably my browser?) 😅

[pjox]

I just merged the tables as suggested 😄 . However I noticed something weird, the train sizes are identical for both the original and deduplicated files ... This is not normal, in general the original files are almost twice as big as the deduplicated ones 🤔

[lhoestq]

Good catch @pjox ! I just checked and this is because the scripts doesn't handle having several blank lines in a row.
Blank lines introduced by deduplication are currently not ignored so we end up with the same number of examples in the dataset as the original version (but with empty examples...)
I fixed that in this commit. I'm re-running the metadata generation for deduplicated configs.

[lhoestq]

I got the new sizes today, will update the dataset_infos.json and the dataset card tomorrow

[cahya-wirawan]

I got the new sizes today, will update the dataset_infos.json and the dataset card tomorrow

great, I just wanted to report that I got error message "NonMatchingSplitsSizesError" when I tried to load one of the oscar dataset.

[lhoestq]

Hi @cahya-wirawan, which configuration of oscar do you have this issue with ?

[lhoestq]

Ok I see you're having this issue because I haven't updated the sizes yet ! I'm opening a PR

I just checked and indeed there's an issue with the deduplicated configurations since the commit I mentioned above.
I'm fixing this by using the new sizes I got yesterday :)

[lhoestq]

I just updated the size in the table @pjox it should be good now :)
I also updated the sizes in the dataset_infos.json in #1868 (merged)

[cahya-wirawan]

Thanks @lhoestq for fixing the issue, it works now

[lhoestq]

Alright this is all good then ! Thanks a lot @pjox

[pjox]

Thank you so much @lhoestq !