Files
foxhunt/crates
jgrusewski cd7c67f541 feat: KAN spline gates replace sigmoid — learned activation per neuron (NUM_WEIGHT_TENSORS 42->50)
8 cubic B-spline basis functions per gate neuron + residual connection.
Spline coefficients initialized to uniform 0.125 (flat 0.5 gate).
kan_gate_combine replaces glu_combine in forward.
kan_gate_backward replaces glu_backward in backward.
9,216 new parameters (4 x AH x 9).

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
2026-04-15 09:13:49 +02:00
..