All terms
Training
Parameter-Efficient Fine-Tuning
Specializing a model by updating only a small fraction of its parameters.
Definition
Parameter-efficient fine-tuning covers techniques that adapt a model while updating only a tiny fraction of its parameters, leaving the base weights frozen. By inserting or training small add-on parameters, methods like LoRA, prefix tuning, and IA3 cut memory and storage dramatically and make fine-tuning large models practical on modest hardware. PEFT is also the name of a Hugging Face library implementing these methods.