Fine-Tuning TranslateGemma-4B for Better Welsh Translations on an H200 GPU(metalglot.com)

1 pointsby metalglothace 2 meses1 comments

Fine-Tuning TranslateGemma-4B for Better Welsh Translations on an H200 GPU

https://metalglot.com/blog/welsh-translategemma-finetuning-guide/

1 comments

Open source repo: https://github.com/grctest/finetuned-gemmatranslate-cy

5% of the fine-tuning took 40 minutes and cost a couple dollars to prove the process works.

Looking forwards to Flash Attention v4 to leave beta, to test fine-tuning performance on a B200 on the cloud, probably a few months away it seems?

What languages would you train TranslateGemma to be able to translate? I was originally thinking about klingon but the available datasets seemed a bit lacking..