ModelRefs / Prefix Tuning — AI Glossary

Prefix Tuning — AI Glossary

A PEFT method prepending trainable virtual token embeddings to each transformer layer's key-value sequence, steering generation without weight updates.

Overview

Prefix tuning (Li & Liang 2021) trains a continuous prompt prepended to the KV sequence of every layer. Unlike prompt tuning (only the input embedding layer), prefix tuning injects learned context deep in the network. More expressive than prompt tuning; less parameter-efficient than LoRA. Used for dialog personalization and style control.

Reference details

Topictraining
Also known asprefix prompt
Last reviewed2026-06-24

Continue your research

Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Prefix Tuning — AI Glossary.

Frequently asked questions

What is Prefix Tuning?

A PEFT method prepending trainable virtual token embeddings to each transformer layer's key-value sequence, steering generation without weight updates.

Is Prefix Tuning the same as prefix prompt?

Yes — prefix prompt are common aliases for Prefix Tuning.

What concepts are related to Prefix Tuning?

Closely related concepts include adapter, prompt tuning, lora.