Block a user
Add opt-in straight-through Gumbel-softmax combine weights to MoE router
Add opt-in straight-through Gumbel-softmax combine weights to MoE router
Add learnable per-expert width and shared temperature to EnergyRouter
Add learnable per-expert width and shared temperature to EnergyRouter
Store a quantile grid instead of a raw reservoir sample in the setup cache
Store a quantile grid instead of a raw reservoir sample in the setup cache