QLoRA 4 bit quantization - Allows to fine tune LLaMa sized models on consumer GPU! twitter.com

QLoRA is a new method of finetuning large models, that's so efficient that it allows for finetuning up to 100 LLaMa sized models a day in a small cluster.