feat: 🎸 upgrade datasets to 2.3.1

[severo]

No description provided.

[severo]

It seems like we have an issue with the gated datasets:

https://github.com/huggingface/datasets-server/runs/6895195217?check_suite_focus=true

E           aiohttp.client_exceptions.ClientResponseError: 401, message='Unauthorized', url=URL('https://huggingface.co/datasets/severo/dummy_gated/resolve/99194748bed3625a941aaf785740df02ca5762c9/data/train-00000-of-00001.parquet')

cc @lhoestq: do you have an idea? Linked to https://github.com/huggingface/datasets/pull/4472/files?

[lhoestq]

This is fixed in 2.3 ;)

It is due to the hub changing the error code for unauthenticated requests to gated repos

[severo]

Hmmm: I get this error with datasets 2.3.0, and still have it in 2.3.1. It's something else, I'll investigate to make a reproducible example

[lhoestq]

I'm getting this error on my side Oo

ValueError: Arrow type extension<arrow.py_extension_type<pyarrow.lib.UnknownExtensionType>> does not have a datasets dtype equivalent.

We were using extension types in the very beginning of the Image feature type, and we dropped it quickly because it was super buggy. You might have to regenerate this dataset

[severo]

yes, I get this too when loading without the streaming mode. I'll create another test dataset.

[severo]

I deployed and also launched a cache refresh for all the datasets.