Simpletransformers

Added

- Added better support for Chinese text in Language Modeling. [taranais](https://github.com/taranais)
- Added `mobilebert` for classification, NER, QA, and Seq2Seq. [flozi00](https://github.com/flozi00)

Fixed

- Fixed incorrect logic when using `early_stopping_metric_minimize`. [djstrong](https://github.com/djstrong)
- Fixed issue with cache directory being created even when `no_cache` is set. [henninglebbaeus](https://github.com/henninglebbaeus)

Changed

- Running loss is now shown next to the tqdm bar (with the tqdm bar description)
- Removed tokenizers and transformers version pins (added earlier to avoid compatibility issues)

Fixed

- Fixed bugs with args not being passed correctly to wandb in the following models:
- `MultiModalClassificationModel`
- `ConvAIModel`
- `Seq2SeqModel`
- `T5Model`
- Fixed bugs in `Seq2SeqModel` and `T5Model` when not using `use_multiprocessed_decoding`.

Changed

- Set `use_multiprocessed_decoding=False` as default for Seq2Seq models to avoid a bug.

Fixed

- Fixed bug where the returned value from `MultiModalClassificationModel.evaluate_model()` was incorrect.

Added

- NER lazy loading support added [Pradhy729](https://github.com/Pradhy729)

Changed

- Added `lazy_loading` attibute to `ClassificationArgs` which now controls whether lazy loading is used.
- Replaced `lazy_loading_header_row` attribute in `ClassificationArgs` with `lazy_loading_start_line`.
- Unnecessary Docs spacing removed [bryant1410](https://github.com/bryant1410)
- Set required tokenizer version to 0.7 until breaking changes are resolved.