Skip to content

Latest commit

 

History

History
43 lines (28 loc) · 1.66 KB

File metadata and controls

43 lines (28 loc) · 1.66 KB

HyperKAN Recovery: Phase A Results

This doc summarizes the Phase A goal: make HyperKAN competitive with Static KAN on the shallow verified benchmark (274 non-terminal test rows; beam width 4; max steps 8).

Baselines (from shallow benchmark)

  • Static KAN beam: 163/274, depth-3: 16/127
  • HyperKAN (original) beam: 148/274, depth-3: 1/127

Recovery Variants (trained + evaluated)

All variants are HyperKAN and use the same dataset split and eval settings; only architecture knobs changed.

Variant Key change Greedy Beam Rescue Depth-3 beam
last_layer_only condition only kan_2 (uniform mix on kan_1) 146/274 156/274 +10 9/127
templates_2 spline_templates: 2 147/274 156/274 +9 9/127
small_hyper hyper_hidden_dim: 64 147/274 162/274 +15 15/127
soft_routing mixture_temperature: 2.0 + mixture_entropy_weight: 0.02 146/274 148/274 +2 1/127

Winner: small_hyper. It nearly matches Static KAN (162 vs 163 overall; 15 vs 16 on depth-3).

Search Temperature Calibration (eval-only)

Applied policy_temperature at beam search time on the small_hyper checkpoint.

policy_temperature Beam Depth-3 beam
0.7 162/274 15/127
1.0 162/274 15/127
1.3 161/274 14/127
1.6 156/274 9/127

Conclusion: temperature is not the missing lever; high temperature hurts.

Commands

Run the recovery sweep (train + greedy/beam eval) inside the toolbox:

toolbox run -c llama-rocm-7.2 bash -c 'cd $REPO && source scripts/toolbox_env.sh && python3 scripts/run_hyperkan_recovery_sweep.py'