How to use nllb1.3b model to fine-tune the English to German bidirectional translation task?

[John6666]

Hmmm…

How to fine-tune a NLLB-200 model for translating a new language

Fine-tuning an NLLB model for a new language

Unfortunately no. I tried to delete and recreate the token IDs as suggested, which initially led to a different ordering of the language token IDs, but even after fixing it I basically ended up with the same situation as before. I also dug around in the tokenizer a bit, but I don’t really understand the root cause of the issue.