Open Models Almanac
Methodsmodel architecturenormalization & residual

specific method · filed under model architecture

Head-wise gated attention

Scales each attention head's output with a sigmoid gate, allowing a head to be suppressed.

The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See normalization & residual for the methods filed beside it.

sources
2
model
1
lab adopt it
1
strongest
core

Documented in

Liquid AI

Evidence

2 quotes from 2 sources, one card per model, each opening with its strongest.

How sources treat it
core 2
↑