Datasets:

cfli

/

bge-full-data

like 1

The default config contains 31 while it should generally contain 3 splits maximum (train/validation/test). If the splits twenty_news_groups, biorxiv_abstract, medrxiv_abstract, arxiv_title, biorxiv_title... are not used to differentiate between training and evaluation, please consider defining configs of this dataset instead. You can find how to define configs instead of splits here: https://huggingface.co/docs/hub/datasets-data-files-configuration