Finetune different language pair on pretrained translation model

My goal is to translate classical Chinese to modern Chinese. There is no such model with same laguage pair on huggingface, but lots of model on Chinese to other languages.
1 I wonder if it can be finetuned? I’m a beginner. And I think maybe tokenizer can be reuse.
2 Is it normal that it takes a long time to train? I train YOLO before and it feels like every epoch is much much longer.
The quetions above arise from the poor performance after several thousands steps and quite a long time. I am not sure it is due to not enough trainning or it just can not be done like this?
Any advice will be appreciated.

I am wondering the same. Is it possible to add a new language pair to the transformer if there isn’t that language pair supported by the trained transformer.