DASH-OPD: Discrepancy-Aware Switching with Hysteresis for On-Policy Distillation
Yuchen Xia, Qianguo Sun, Chao Song, Junlong Wu, Yiyan Qi, Yunjian Xu
Abstract
On-policy distillation (OPD) trains student models on their own rollouts to reduce exposure bias. However, in multi-turn agent scenarios, early student errors can lead a trajectory away from the teacher's familiar domain. Existing curriculum learning methods regulate how much teacher support is used according to training progress, but cannot determine when it is needed. In light of this, we propose DASH-OPD, Discrepancy-Aware Switching with Hysteresis for OPD, the first agentic OPD method that can switch executors adaptively and bidirectionally. On each turn, DASH-OPD calculates a mean log-probability ratio between the two executors over action tokens as their discrepancy. Student-to-teacher ratios on student turns form drift signals, while teacher-to-student ratios on teacher turns form recovery signals. These signals are normalized and accumulated over multiple turns into drift and recovery evidence. DASH-OPD switches executors when the evidence exceeds its corresponding switching threshold. This multi-turn accumulation makes the switching hysteretic, preventing high-frequency switches caused by transient fluctuations. Across WebShop, ALFWorld, and ScienceWorld at two student-model scales, DASH-OPD outperforms five baselines in all 14 task-performance comparisons while yielding the shortest trajectories in nine of ten turn-count comparisons, offering the strongest overall performance-efficiency trade-off. This paper is a work in progress. Code, training logs, and model checkpoints will be released later.
Topics
Classified with taxonomy v2 on Sat, 5 Sept 2026.