single mention · model architecture
SWA-1024 (3:1 Interleaving)
Sliding-window attention with a stated window size of 1,024, interleaved in a 3:1 pattern.
The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See sliding window attention for the methods filed beside it.
Also called + SWA-1024 (interleaved 3:1).
Filed under model architecture › token mixer › softmax attention › sliding window attention