Adding the fast tokenizer for inference to be able to load (slow->fast fe7b148 Narsil HF staff commited on Feb 12, 2021