aiqbench analysis
DASH-OPD addresses exposure bias in on-policy distillation by introducing hysteresis mechanisms that adaptively switch between student and teacher executors based on accumulated discrepancy evidence. This approach prevents high-frequency switching during transient fluctuations, improving training efficiency and stability for multi-turn agentic tasks without relying on fixed curriculum schedules. The method demonstrates superior performance over existing baselines in complex environments like ALFWorld while remaining a work-in-progress pending full release of code and models.
DASH-OPD: Discrepancy-Aware Switching with Hysteresis for On-Policy Distillation
arXiv:2607.29078v1 Announce Type: new 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, a new 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. On ALFWorld, DASH-OPD outperforms all the baselines and demonstrates superior training and deployment efficiency. This paper is a work in progress. Code, training logs, and model checkpoints will be released later.
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, a new 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. On ALFWorld, DASH-OPD outperforms all the baselines and demonstrates superior training and deployment efficiency. This paper is a work in progress. Code, training logs, and model checkpoints will be released later.
Machine Learning (cs.LG)
arXiv:2607.29078 [cs.LG]
(or arXiv:2607.29078v1 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2607.29078 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 280 words
Related coverage
[AINews] Reality Checks on AI News (Yegge shuts down Gas Town, Databricks’ +60% Astra cost)
Latent Space · Sep 17, 2026
DASH: Divergence-Adaptive Supervision Horizons for On-Policy Self-Distillation of Reasoning Models
arXiv · Aug 7, 2026
Beyond In-Distribution Preservation: Recovering Generalization in Quantized VLAs via Vulnerability-Oriented Tuning
arXiv · Oct 6, 2026
CV-OPD: Contrastive Vision On-policy Distillation
arXiv · Oct 12, 2026