Use /info as the source for configs and splits? about dataset-viewer HOT 4 CLOSED

Also: why not use get_dataset_config_names for configs/ ?

https://mobile.twitter.com/_lewtun/status/1430550924980797446

from dataset-viewer.

Comparing the configs and splits with the data provided by info, we see a lot of mismatches: 20% of the datasets show some mismatch.

Sometimes (https://github.com/huggingface/datasets/blob/master/datasets/biomrc/biomrc.py -> while info provides a "plain_text" config: https://github.com/huggingface/datasets/blob/master/datasets/biomrc/dataset_infos.json) the info is wrong, sometimes (https://github.com/huggingface/datasets/blame/master/datasets/europarl_bilingual/europarl_bilingual.py#L106) the configs endpoint is wrong... We should add this checks to the "e2e" dashboard to regularly check the quality of datasets

from dataset-viewer.

Also: note that info is not available for datasets that only contain data files, like https://huggingface.co/datasets/lhoestq/demo1/tree/main. As we are pushing towards this kind of repositories (see https://huggingface.co/docs/datasets/loading_datasets.html#from-a-community-dataset-on-the-hugging-face-hub), I imagine that we should not rely on info.

from dataset-viewer.

Note that our implementation is the same as get_dataset_config_names and get_dataset_infos. Closing for now.

from dataset-viewer.