- Python 3.10 support [luketudge](https://github.com/luketudge)
Changed
- Removed usage of deprecated function `batch_encode_plus` [whr778](https://github.com/whr778)
0.63.8
0.63.7
Added
- Added support for LayoutLMV2 and RemBERT [whr778](https://github.com/whr778)
Fixed
- Fixed issue with ner_utils lazy_loading_start_line not being set correctly. [whr778](https://github.com/whr778) - Lazy loading bug fixes [sainttttt](https://github.com/sainttttt) - Fixed seq2seq and T5 preprocessing [MichelBartels](https://github.com/MichelBartels)
0.63.6
0.63.5
Added
- Added support for selecting FAISS index type with `RetrievalModel`.
0.63.4
Added
- Added support for individual training of context/query encoders in `RetrievalModel`.
Fixed
- Fixed a bug for pre-tokenized input to ONNX models. [whr778](https://github.com/whr778) - BigBird bugfix where training dataset samples were being truncated at 510 token. [whr778](https://github.com/whr778) - Fixed bug when FP16 is not used with `RetrievalModel`. [tiena2cva](https://github.com/tiena2cva) - Fixed bug in auto model for `QuestionAnsweringModel`. [lekhang4497](https://github.com/lekhang4497) - Fixed bug where wrong predictions weren't returned in `ClassificationModel` [jinschoi](https://github.com/jinschoi)