Show HN: Phase Router – capacity-aware routing for MoE

2 comments

• TSltd 2 days ago

  Reducing dropped tokens could also improve model training by reducing gradient noise

• TSltd 3 days ago

  [dead]