- Fixed bug in binary classification evaluation when data only contains one label. - Fixed typo in readme.
Changed
- Cache dir is no longer created when `no_cache` is used.
0.23.3
0.23.2
0.23.1
0.23.00.23.0
Added
- Added support for training custom tokenizers. - Added improved support for training language models from scratch. - Added improved support for resuming training in classification, NER, and QnA tasks.