Glossary
What is IA3? Definition and meaning
Adapts a model by learning vectors that scale keys, values and feed-forward activations, adding very few parameters.
Definition
IA3: Adapts a model by learning vectors that scale keys, values and feed-forward activations, adding very few parameters.
In short: Learning small rescaling vectors on activations.
When it is used
Few-shot adaptation with an extremely small parameter budget.
Learn more
The full guide to IA3 explains how it works, when to use it and how it compares with the alternatives, with examples. The glossary lists every term on the site from A to Z.
Previous term: Hyperparameter tuning · Next term: IBM TrueNorth