arXiv Computation and Language

Learning from Think-Mode Advantage via On-Policy Distillation

arXiv Computation and Language
Sep 18

What Does Privileged Information Add to On-Policy Self-Distillation?

The paper investigates how privileged information—such as a teacher’s full solution or reasoning trace—affects on‑policy self‑distillation (OPSD) in language models. Using the AMPLE‑Math benchmark, the authors compare distillation with and without extra teacher views, finding that reference‑free distillation explains most gains for Qwen3‑1.7B, while additional references provide modest benefits, especially for polished solutions. The study also shows that the impact of privileged data depends on the student’s training regime and that altering token‑level supervision can leave student behavior largely unchanged.

By XiuYu Zhang, Wei Chow, Junfeng Fang, Zhenkai Liang, Tat-Seng Chua
arXiv AI
Jul 3

Purified OPSD: On-Policy Self-Distillation Without Losing How to Think

arXiv:2607. 02234v1 Announce Type: new Abstract: On-policy self-distillation (OPSD) has emerged as a promising paradigm for improving LLM reasoning, where a privileged teacher with access to reference solutions provides token-level supervision on the student's own generated trajectories.

By Zhanming Shen, Jintao Tong, Shaotian Yan, Chen Shen, Hao Chen, Wentao Ye, Xiaomeng Hu, Rui Miao, Haobo Wang, Junbo Zhao, Gang Chen, Jieping Ye
arXiv Computation and Language
Oct 1

Diagnosing On-Policy Self-Distillation for Reasoning Language Models

The paper investigates on‑policy self‑distillation (OPSD) as a method to enhance reasoning in language models, focusing on mathematical reasoning across models from 0.6B to 8B parameters. Through controlled experiments and token‑level analysis, the authors find that OPSD’s effectiveness depends on alignment between the teacher’s reasoning mode and the full teacher prefix, rather than on privileged semantics alone. They observe that OPSD only improves reasoning in limited compatibility regimes, while often causing length growth, degradation, or behavioral collapse, and that the teacher’s signal is unstable and not predictive of downstream performance.

By Yang Li, Gongle Xue, Yuheng Yuan, Yijia Guo, Shizhe Zhang, Liwen Hu, Lei Ma
arXiv AI
Oct 28

From Solver Trajectories to Teaching Trajectories: Cognition-Aligned Reasoning Distillation

The paper introduces Cognition-Aligned Reasoning Distillation (CARD), an RL-based method that converts solver trajectories from large reasoning models into teaching trajectories better suited for smaller student models. CARD uses two reward signals—a Distribution Alignment Reward and a Logical Coherence Reward—to align the teacher’s outputs with the student’s cognitive capabilities and reduce misleading reasoning steps. Experiments show that CARD improves both in-domain and out-of-distribution performance compared to distillation from raw teacher trajectories.

By Yashuo Luo, Qianren Mao, Ziqi Qin, Junnan Liu, Haitao Li, Qili Zhang, Yuan Gao, Zhijun Chen, Xu Wang
arXiv Machine Learning
Oct 16

Waypoint-OPD: Transferring RL-Induced Policy Shifts via Gated On-Policy Distillation

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
arXiv AI
6d ago

Bootstrapped On-Policy Distillation

Bootstrapped On-Policy Distillation (BOPD) is a two‑stage framework that first uses on‑policy self‑distillation (OPSD) to capture task‑relevant knowledge from reference‑conditioned supervision into an intermediate model, and then distills this intermediate model into a fresh copy of the initial model. The approach addresses the degradation observed in OPSD during continued training, which is caused by a growing mismatch between the evolving student and the frozen teacher. Experiments demonstrate that BOPD improves mathematical reasoning and code generation beyond both the initial models and the constructed OPSD teachers.

By Yue Wu, Xueyu Zhou, Yangrong Hu, Jian Huang