Data & AI

Mixture of Experts (MoE) Inference Routing

Data & AI Pipelines diagram · made in Skemio

Architecture diagram showing token embeddings flowing to a Top-2 router, dispatching to selected Expert FFN blocks, and merging at a weighted combiner.
Architecture diagram showing token embeddings flowing to a Top-2 router, dispatching to selected Expert FFN blocks, and merging at a weighted combiner.
Open it in SkemioMake your own

What it shows

  • Gating networks dynamically compute Top-K routing probabilities per token.
  • Tokens route only to specialized feed-forward sub-networks (experts).
  • A weighted combiner merges expert outputs without dense compute overhead.

The takeaway

MoE expands total parameter capacity while keeping active compute cost per token flat.

More data & ai pipelines diagrams