It is a NLLB-200-600M model with extended vocabulary, fine-tuned for translating between Tyvan and Russian languages using the dataset from https://tyvan.ru.

Here is a post about how it was trained.