transformers

c4d78f01 - Fix tokenizer saving during training with `Trainer` (#12806)

Fix tokenizer saving during training with `Trainer` (#12806) * add test in trainer and test tokenizer saving wi th trainer * quality * reverse trainer changes * replace test in test_trainer by a test for all the tokenizers * format * add can_save_slow_tokenizer attribute to all tokenizers * fix Herbert * format * Change comment in error * add comments and a new assert * Update src/transformers/models/albert/tokenization_albert_fast.py Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com> * change ValueError barthez * change ValueError BigBird * change ValueError Camembert * change ValueError Mbart50 * change ValueError Pegasus * change ValueError ReFormer * change ValueError T5 * change ValueError RoBERTa * XLNET fast * Update tests/test_tokenization_common.py Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com> * change `assert` into `self.assertIn` * format Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>