1% of Tokens Can Be Enough: On Gradient Estimation in On-Policy Distillation
Sparse on-policy distillation (OPD) allocates teacher supervision to a small subset of tokens in student-generated trajectories. However, useful teacher guidance can yield a noisy update when its gradient is estimated from a sampled next token. We study this estimation problem at a fixed prefix in information geometry and propose an information-efficiency ratio (IER) based on a
Reading in Plain English mode — toggle below to switch
Sparse on-policy distillation (OPD) allocates teacher supervision to a small subset of tokens in student-generated trajectories. However, useful teacher guidance can yield a noisy update when its gradient is estimated from a sampled next token. We study this estimation problem at a fixed prefix in information geometry and propose an information-efficiency ratio (IER) based on a
This research explores sparse on-policy distillation, a method to efficiently train models by focusing teacher supervision on a small, yet critical, subset of tokens. For ML engineers, understanding this approach can lead to significant cost savings and faster iteration cycles in model development, particularly for complex reasoning tasks.