PEFT
PEFT — Parameter-Efficient Fine-Tuning, a family of adaptation methods that update a small subset or auxiliary set of parameters instead of retraining an entire model.
Machine-Learning Context
Parameter-Efficient Fine-Tuning is a family of adaptation methods that trains only a small subset of parameters or adds compact trainable components. LoRA, adapters, and prompt- or prefix-based methods can reduce accelerator memory and checkpoint size relative to full-model fine-tuning.
Adaptation Boundary
PEFT is not one algorithm and does not guarantee that every behavior of a base model can be reshaped reliably with a small parameter budget. Method choice depends on the target task and model.