single mention · model architecture
Search-Based Sliding-Window Attention Pattern
Selects layers for sliding-window conversion through a search-based adaptation while retaining full attention in the other layers.
The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See hybrid layer stacking for the methods filed beside it.
Also called SWA Pattern (Search-Based).
Filed under model architecture › token mixer › hybrid layer stacking