Skip to content
Technical GlossaryGenerative AI and LLM

Prefix Tuning

A PEFT technique that steers the model’s internal attention behavior through small learnable prefix representations.

Prefix tuning learns internal context-like vectors to guide task behavior without changing most of the model weights. It is an interesting alternative in scenarios where parameter efficiency is important. It can be seen as one of the approaches that bridges prompting and fine-tuning.