Open Models Almanac
Methodsmodel architecturetoken mixersoftmax attentiongrouped-query attention

single mention · model architecture

Block-sparse grouped-query attention

A block-sparse variant of grouped-query attention that replaces the full attention stack.

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

Also called block-sparse GQA.

Filed under model architecture › token mixer › softmax attention › grouped-query attention