Router Network
Another name for the gating module — it routes each token to its chosen experts.
Think of it like
Like a mailroom sorter reading each envelope and dropping it in the right department’s bin.
Example
A model card might note the router sends each token to two of its many experts.
How it actually works
“Router” and “gating network” usually point at the same thing: the component that decides which experts process a token. The word “router” emphasizes the dispatch, “gating” the weighting. Both must balance load across experts and route consistently, or the model wastes capacity and trains poorly.
For product teams
Same idea as the gating network, described from the dispatch angle.
For engineers
The MoE component computing per-token expert assignments and dispatch weights.
Related
- Gating Network — The other common name for it.
- Expert Routing — The routing decision itself.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome