AI Glossary Term

What does “Parameter-Efficient Fine-Tuning (PEFT)” mean?

Definition

Methods that adapt models by training a small subset of added parameters.

Browse the full AI Glossary Explore all free AI guides

How Parameter-Efficient Fine-Tuning (PEFT) works

Example

A small team wants a model that writes in their company's style. With PEFT they train only a tiny add-on, using a fraction of the memory that full fine-tuning would need, and keep the original model untouched.

Why it matters

PEFT cuts the cost of customizing large models, which opens fine-tuning to teams without large budgets.

Common questions

Which methods count as PEFT?

LoRA, adapters, prefix or prompt tuning, and similar methods that update a small number of parameters.

Is quality lower than full fine-tuning?

Often comparable for many tasks, though not always.

Can one base model serve many tasks?

Yes. You can keep one base and switch small task-specific modules.

Fine-TuningContinuing training on domain-specific data to adapt a pre-trained model to a specific task.LoRA (Low-Rank Adaptation)A parameter-efficient fine-tuning method that adds low-rank adapter matrices.Post-trainingTraining steps applied after pretraining, such as instruction tuning, preference optimization, and safety tuning.Instruction TuningFine-tuning a model on instruction-response pairs to improve task following.Learning RateA training hyperparameter controlling how much parameters change each update step.DPO (Direct Preference Optimization)A training method that fine-tunes models directly on preference pairs without needing a separate reward model.

Learn more in our free guides

Fine-Tuning Hard Parameter Sharing in Multi-Task Networks Gumbel-Softmax and Reparameterization Rejection Sampling Fine-Tuning

Test yourself

Fine-Tuning Quiz Hard Parameter Sharing in Multi-Task Networks Quiz Gumbel-Softmax and Reparameterization Quiz