AI Inference & Serving

Mixture of Experts

In this inference context, sparse MoE routes tokens through selected expert components rather than activating every expert for every token. These experts are neural-network components, not independent people or chat agents. The broader mixture-of-experts family also includes other designs.

Also called: MoE

Reviewed

Sources

Free complete lesson

This definition and the complete practical lesson are free. The interactive reader loads the lesson examples without a membership.

Find the free complete lesson

Explore all dictionary definitions