Fine-tuning TrOCR on new language

It seems that the error may occur due to the tokenizer settings.