single mention · model architecture
Coarse-grained MoE configuration
A configuration using fewer, wider experts, exemplified by six selected from 256 experts.
The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See fine-grained experts for the methods filed beside it.
Also called coarser-grained configuration.
Filed under model architecture › channel mixer › mixture of experts › fine-grained experts