List your company

Control Smoothness on Rollouts (Held-out)

1.46Action RateNo-smoothness baseline
0.50320.751611.2484Jul 6, 2026

Evaluation results

MethodLinks
No-smoothness baseline2026.071.46
RL fine-tuning2026.071.03
Multi-teacher student2026.070.66
Single-teacher student2026.070.56
SONIC-Base2026.070.54