PEFT

PEFT is Hugging Face's library for parameter-efficient fine-tuning methods like LoRA, QLoRA, and prompt tuning, letting you adapt large models by training a tiny fraction of parameters on modest hardware.