Expert Routing
The traffic cop inside a mixture-of-experts model that sends each token to just a few specialized sub-networks.
Think of it like
Like a hospital triage nurse deciding which specialist each patient should see, so nobody wastes the whole staff on one case.
Example
In a mixture-of-experts model, routing might send a coding token to two “code” experts and skip the other sixty.
How it actually works
Instead of every token passing through every parameter, a router picks a small subset of “experts” per token, so the model can be huge in total but cheap per token. Getting the routing right is the hard part — bad routing overloads some experts and starves others. It’s the mechanism that makes sparse giant models practical to run.
For product teams
Why some “huge” models are still affordable to serve — most of them stays idle per query.
For engineers
The learned assignment of tokens to a top-k subset of experts in an MoE layer.
Related
- Mixture of Experts — The model type it lives inside.
- Gating Network — The small network that makes the choice.
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