How to get the 8/4 bits model?

#8
by ej2 - opened

Which part I should modify so that my finetune is normal at 32/16 bits, but when saving the model, it is saved as 4/8 bits?
Should I change the “finetune.sh” or “train. py”?
I also ask you in https://github.com/haotian-liu/LLaVA/issues/858

Sign up or log in to comment