Learn new & fine-tune existing language pairs in NLLB

[molokanov50]

Hello team,

Is it possible to:

• fine-tune any language pair in a pretrained NLLB model;
• learn an NLLB model for a new language pair, without the full retraining of existing lang pairs?

If yes, where to find infos about dataset size, dataset format, hardware necessary for learning?

[AriNubar]

I am also interested in this. How can one keep the pretrained language knowledge while introducing a new language?