Finetune LLaMa2 for any language

We've released convenience scripts to fine-tune LLaMa2 to any language (that isn't English) using (Q)LoRA. Total training cost per language is under $1. We've already released a few datasets and models to play around with, more to come.

https://github.com/UnderstandLingBV/LLaMa2lang

Few results from the Dutch 7B one:

Q: Wat is de hoofdstad van Nederland?

A: Amsterdam

Q: In welke provincie ligt die stad?

A: In de provincie Noord-Holland.

Q: Wie is de minister-president van Nederland?

A: Mark Rutte is sinds 2010 …

cost datasets english language languagetechnology llama2 lora per scripts total training van