Open Models Almanac
Methodsmodel architecturetoken mixer

single mention · model architecture

Kernel-4 convolutions in decoder blocks

Decoder layers apply kernel-size-4 convolutions after key and value projections and to attention and MLP branch outputs.

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

Filed under model architecture › token mixer