Open Models Almanac
Methodsmodel architecturetoken mixersparse attention

single mention · model architecture

Sparse attention training from scratch without dense-attention warmup

A training procedure that starts sparse attention at 64K sequence length without dense-attention warmup.

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

Filed under model architecture › token mixer › sparse attention