The paper introduces Δ‑MOPD, a multi‑teacher on‑policy distillation method that transfers each teacher’s logit shift relative to its base policy, re‑anchored at the student’s frozen initialization. Δ‑MOPD is evaluated in two settings: common‑domain composition, where multiple teachers score a single student rollout, and routed‑domain distillation, where prompts are assigned to specialist teachers. Across experiments, Δ‑MOPD outperforms traditional endpoint supervision, especially when teacher signals are combined at a state, achieving higher accuracy and reducing performance gaps in phased routing scenarios.
By Hejian Sang, Zhengze Zhou, Shayan Mohajer Hamidi, Xiaomin Li, Rohit Jain, Alborz Geramifard
arXiv:2609.14636v1 Announce Type: new
Abstract: On-policy distillation (OPD) has become a standard approach for transferring capabilities from large teachers to compact students. Its cost, however, i...
By Zhiyu Gui, Kexin Huang, Jia Guo, Junkang Wu, Zihao Wang, Zhiqiang Zhang, Jun Zhou, Jiancan Wu, Xiang Wang
MOPD‑Router rethinks teacher routing in multi‑teacher on‑policy distillation by routing supervision over the full teacher pool at each token, eliminating the need for prompt‑level domain labels or a separate routing model. The framework offers a plug‑in interface for various metrics, and introduces ExpertAlign, which scores teachers based on how well their corrections reflect their specialized post‑training knowledge. Experiments on both unlabeled and domain‑labeled mixtures show that ExpertAlign outperforms existing methods, improving overall scores by up to 12.3% on unlabeled data and 7.8% on domain‑labeled data.
whyItMatters":"Token‑level routing enables the use of complementary supervision across domains without relying on domain labels, leading to significant performance gains in multi‑teacher distillation settings."
By Tianze Xu, Yanzhao Zheng, Zhentao Zhang, Yuanqiang Yu, Chao Ma, Jihuai Zhu, Lelun Wu, Lyumanshan Ye, Pengfei Liu, Baohua Dong, Hangcheng Zhu, Ruohui Huang, Gang Yu
The paper introduces Selective Supervision for Direct-OPD (S$^2$D-OPD), a refinement of Direct On-Policy Distillation that filters out states where the teacher’s policy change is minimal, as measured by the teacher‑reference Jensen‑Shannon divergence. By masking low‑divergence states and keeping only the top 10% of states per response, S$^2$D-OPD improves held‑out accuracy on AIME and HMMT benchmarks across multiple teacher‑student pairs without additional forward passes.
By Yibo Zhao, Zixuan Yang, Yunshi Lan, Xiang Li
The paper proposes a reward‑density principle for language‑model post‑training: sparse sequence‑level rewards are best for exploring and discovering new behaviors in a strong teacher, while dense token‑level teacher supervision is optimal for compressing that behavior into a smaller deployment model. The authors validate this rule with a four‑stage workflow—teacher RL, forward‑KL warmup, on‑policy distillation, and optional post‑bridge student RL—showing that an RL‑improved 8B teacher distilled through a dense bridge outperforms direct GRPO on a 1.7B student across math and code benchmarks. Ablation studies confirm each stage’s contribution, and the approach generalizes to Llama and Qwen models, offering a practical guideline for allocating scarce labeled data.
By Yuanda Xu, Hejian Sang, Zhengze Zhou, Ran He, Zhipeng Wang, Alborz Geramifard
arXiv:2608. 09447v1 Announce Type: cross Abstract: On-policy distillation (OPD) aligns a student with a teacher on trajectories sampled from the student itself, reducing the train-test state mismatch of offline distillation.
By Zehao Chen, Gongxun Li, Tianxiang Ai, Yifei Li, Zixuan Huang, Wang Zhou, Tao Huang, Fuzhen Zhuang, Xianglong Liu, Jianxin Li, Deqing Wang, Yikun Ban
arXiv:2609.36484v1 Announce Type: new
Abstract: On-policy distillation (OPD) trains a student to match the teacher's next-token distributions on the student's own trajectories and has yielded substan...
By Hao Li, MeiJia Chen, Weijie Ren, Donghan Li, Zijun Tian, Jingchun Huang, Naibo Wang
Waypoint-OPD is a new on-policy distillation method that leverages an intermediate teacher checkpoint to compare early and late policy shifts in reinforcement learning. By gating the endpoint shift based on the directional agreement between these early and late shifts, the method suppresses conflicting updates and requires only one additional frozen-teacher scoring pass. Experiments with Qwen3-4B teachers and larger Qwen3 student models in mathematical and code reasoning tasks show that Waypoint-OPD consistently outperforms endpoint-only transfer across all six domain–student settings.
By Bingxi Hou, Guochao Jiang, Guofeng Quan, Weiqing Li, Wenfeng Feng, Guohua Liu, Yuewei Zhang
The paper introduces D$^3$-MOPD, a dynamic domain scheduling method for multi-teacher on‑policy distillation. It adapts the domain mixture during training by monitoring each domain’s reverse‑KL trajectory, thereby allocating more compute to slower‑converging domains and less to those that plateau early. Experiments on a Qwen3.6‑35B‑A3B student show that D$^3$-MOPD closes 97% of the student‑to‑teacher performance gap, matches peak performance with roughly three times fewer rollout steps, and outperforms specialist teachers on most benchmarks.
By Zechen Sun, Zhiwei Zhang, Fei Zhao, Juntao Li, Mu Chuan, Huayu Deng, Guojian Zhan, Wenliang Chen, Yao Hu, Min Zhang
The paper investigates whether on‑policy distillation (OPD) is always advantageous for transferring knowledge from a teacher to a student model. It introduces Semi‑OPD, which uses offline rollouts from the initial student, and demonstrates that across 17 teacher‑student pairs (1.5B–235B parameters) Semi‑OPD outperforms OPD in 14 cases, achieving up to +13.6% accuracy and an 11.4× speedup. The authors show that the benefit of OPD depends on the alignment between teacher and student, measured by an output‑token overlap ratio, and argue that effective distillation requires on‑policy behavior relative to both models.
By Siyan Zhao, Yonggan Fu, Jindong Jiang, Shih-Yang Liu, Song Bian, Byung-Kwan Lee, Sharath Turuvekere Sreenivas, Wenliang Dai, Hanrong Ye, Aditya Grover, Pavlo Molchanov
arXiv:2607. 26246v1 Announce Type: new Abstract: On-policy distillation (OPD), which aligns a student with the teacher's token-level distribution on the student's own rollouts, is an effective paradigm for transferring capabilities across LLMs.
By Fangxu Yu, Zinan Lin, Xiaodong Liu, Weijia Xu, Michael Xu, Tianyi Zhou, Jianfeng Gao
The study compares supervised fine‑tuning (SFT) and reinforcement learning (RL) as methods for training teacher models that are later merged into a single student via on‑policy distillation (OPD). Across agentic, reasoning, and perception tasks, RL‑trained teachers produce stronger students and better recover teacher performance gains than SFT teachers, both in single‑teacher and multi‑teacher OPD settings. Analysis shows RL teachers deviate less from the shared initialization, leading to faster and higher overlap in teacher–student top‑K predictions, which appears to facilitate more effective OPD learning.
By Jingyuan Huang, Zuming Huang, Yucheng Shi, Zhongzhi Li, Xiaoming Zhai, Wei Chu, Ninghao Liu