single mention · model architecture
MoE with expert sinks
A shared-expert arrangement in which shared experts contribute to routing-score computation but are excluded from the routed top-k candidate pool.
The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See shared experts for the methods filed beside it.
Also called Mixture-of-Experts with expert sinks.
Filed under model architecture › channel mixer › mixture of experts › shared experts