single mention · model architecture
Latent-dimension reduction with increased expert count and active experts
An implementation detail that reduces routed-expert input dimensionality and reinvests the savings in nonlinear capacity and expert diversity.
The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See latent mixture of experts for the methods filed beside it.
Filed under model architecture › channel mixer › mixture of experts › latent mixture of experts