undefined | Better HN

0 pointsbrucethemoose22y ago0 comments

There will be finetunes for other languages just like LLaMAv1

0 comments

2 comments · 1 top-level

1024core2y ago· 1 in thread

How can you finetune for a new language? Aren't the tokens baked in by the time the model is done training?

Apparently not. shrug

The backend does sometimes need a new tokenizer, depending on how its implemented.

j / k navigate · click thread line to collapse