01Router Saturation

02Expert Activation Norms

Per-layer

Browse all layers (scroll)

03Coactivation Matrices

Selected layer

Browse all layers (scroll)

04Expert Routing per-layer, frame-stepped

Compare up to 3 layers side-by-side. Use Play/Pause, ◀/▶, the slider, or ←/→ to step through training checkpoints.

Aggregate routing GIF (first/mid/last)

05Raw JSON