Make student models surpass teachers through reward extrapolation in distillation.
On-policy distillation (OPD), which aligns the student with the teacher's logit distribution on student-generated trajectories, has demonstrated strong empirical gains in improving student performance and often outperforms off-policy distillation and reinforcement learning (RL) paradigms. In this work, we first theoretically show that OPD is a special case of dense KL-constrained RL where the reward function and the KL regularization are always weighted equally and the reference model can by any model. Then, we propose the Generalized On-Policy Distillation (G-OPD) framework, which extends the standard OPD objective by introducing a flexible reference model and a reward scaling factor that controls the relative weight of the reward term against the KL regularization. Through comprehensive experiments on math reasoning and code generation tasks, we derive two novel insights: (1) Setting the reward scaling factor to be greater than 1 (i.e., reward extrapolation), which we term ExOPD, consistently improves over standard OPD across a range of teacher-student size pairings. In particular, in the setting where we merge the knowledge from different domain experts, obtained by applying domain-specific RL to the same student model, back into the original student, ExOPD enables the student to even surpass the teacher's performance boundary and outperform the domain teachers. (2) Building on ExOPD, we further find that in the strong-to-weak distillation setting (i.e., distilling a smaller student from a larger teacher), performing reward correction by choosing the reference model as the teacher's base model before RL yields a more accurate reward signal and further improves distillation performance. However, this choice assumes access to the teacher's pre-RL variant and incurs more computational overhead. We hope our work offers new insights for future research on OPD.
On-policy distillation is equivalent to KL-constrained reinforcement learning with implicit rewards.
Reward extrapolation with lambda > 1 enables students to surpass teacher performance.
Generalized On-Policy Distillation unifies knowledge distillation and reinforcement learning frameworks.
Reference model selection significantly impacts extrapolation effectiveness but lacks clear guidance.
Approach requires student to generate trajectories, increasing computational cost versus off-policy methods.
Guobin Shen, Chenxiao Zhao +3
Training stability remains a central challenge in reinforcement learning (RL) for large language models (LLMs). Policy staleness, asynchronous training, and mismatches between training and inference e...
Zhixia Zhang, Zixuan Huang +8
We introduce Heterogeneous Agent Collaborative Reinforcement Learning (HACRL), a new learning paradigm that addresses the inefficiencies of isolated on-policy optimization. HACRL enables collaborative...
Jiyuan Wang, Chunyu Lin +9
Leveraging the priors of 2D diffusion models for 3D editing has emerged as a promising paradigm. However, maintaining multi-view consistency in edited results remains challenging, and the extreme scar...
Fengkai Yang, Zherui Chen +11
Reinforcement Learning from Verifier Rewards (RLVR) has emerged as a widely used approach for post-training large language models on reasoning tasks, with group-based methods such as GRPO and its vari...
Yanqi Dai, Yuxiang Ji +4
Reinforcement Learning with Verifiable Rewards (RLVR) offers a robust mechanism for enhancing mathematical reasoning in large models. However, we identify a systematic lack of emphasis on more challen...
