Skip to main content

Simulation run on this device. Scenario values are not sent anywhere — the engine is fully local.

ai game · free

MoE Routing Arcade

Route tokens across experts under a capacity constraint and keep the load even.

Can you balance the experts before one melts down?

Engine 1.0.0 · game
464
2
1.25
0.4
Imbalance (max/min)
16.0×
Overloaded experts
3 / 16
Dropped capacity
631 tok·slots
Slots / expert
160top-2

Hot experts are truncating — raise capacity factor or flatten routing with aux loss.

E=16 top-2 cap×1.25 skew=0.40
hottest 606 vs coldest 38

Method

  • Zipf-skewed expert demand under top-k and capacity factor; overload and dropped token·slots are the teaching signal.
  • Numbers are teaching approximations, not production capacity guarantees.