single mention · model architecture
Full Attention Residuals
Uses layer-specific learned pseudo-queries to attend over earlier layer representations.
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.
Filed under model architecture › normalization & residual