Open Models Almanac
Methodsmodel architecturechannel mixermixture of experts

single mention · model architecture

Dispatch recomputation

A memory-saving backward-pass technique that recomputes dispatch to recover the group-GEMM input.

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

Also called recomputing dispatch.

Filed under model architecture › channel mixer › mixture of experts