stabilityai

/

StableBeluga2

like 883

How can we finetune this? And could be the resources requirement for it.

you fine tune it like a normal llm and keep in mind https://twitter.com/moinnadeem/status/1681393075367841792

and you will need 4,400 gigs of vram (with nothing to lower vram )

and you will need 4,400 gigs of vram (with nothing to lower vram )

But why? I thought LLAMA 2 can be fine-tuned with single A-100

and you will need 4,400 gigs of vram (with nothing to lower vram )

But why? I thought LLAMA 2 can be fine-tuned with single A-100

The 7b model yeah and the 13b model yeah with qlora but 70b model can't fine-tuned on a single gpu even with qlora