# What is Prompt tuning? Definition and meaning

Trains a small set of embedding vectors that are prepended to the input, with the model frozen.

## Definition

**Prompt tuning**: Trains a small set of embedding vectors that are prepended to the input, with the model frozen.

In short: Learning soft prompt embeddings.

## When it is used

The cheapest form of task adaptation for large models.

## Learn more

The full guide to **[Prompt tuning](/tools/prompt-tuning)** explains how it works, when to use it and how it compares with the alternatives, with examples. The [glossary](/glossary) lists every term on the site from A to Z.

Previous term: [Prompt engineering](/glossary/prompt-engineering) · Next term: [Prompt-injection defense](/glossary/prompt-injection-defense)
