This dataset is copied from CLUE with certain modifications. The paper of CLUE is OCNLI.
The modifications are:
Transform json file to csv file.
Encoding in UTF-8.
Remove data entries whose label value is '-'.
Replace label values, 'neutral' to 1, 'entailment' to 0, and 'contradiction' to 2.
The train file uses trank.3k
