3 ms·
QLoRA 4-bit finetuning of LLMs
- kashifr 3y agoAn efficient finetuning approach that reduces memory usage enough to finetune a 65B parameter model on a single 48GB GPU while preserving full 16-bit finetuning task performance!
loading thread…
3 ms·