paper-with-me

홈 › Papers

On the Geometry of On-Policy Distillation

2026-06-05 · Zhennan Shen, Yanshu Li, Qingyu Yin, Chak Tou Leong, Zhilin Wang, Yanxu Chen, Rongduo Han, Sunbowen Lee, Yi R. Fung arxiv

On-policy distillation (OPD) is increasingly used to improve large language model reasoning, but its training dynamics remain poorly understood. We characterize the trajectory of OPD updates in parameter space and compare it with supervised fine-tuning (SFT) and reinforcement learning with verifiable rewards (RLVR). A suite of parameter-space diagnostics consistently places OPD in a relaxed off-principal regime: compared with SFT, its updates affect fewer weights and avoid principal directions more strongly, while compared with RLVR, they remain less tightly constrained. Beyond this static localization, OPD exhibits subspace locking: its cumulative updates rapidly enter a narrow low-dimensional channel. Constraining training to the update subspace formed early in training preserves OPD performance but substantially degrades SFT, indicating that the locked subspace is functionally sufficient for OPD. Control experiments further show that sparsifying the update tokens and shifting rollout generation off-policy preserve the rank dynamics, whereas mixing the OPD objective with RLVR changes them. Overall, these results suggest that OPD is not merely an intermediate point between SFT and RLVR, but induces its own update geometry in parameter space.

📄 PDF Abstract BibTeX arXiv:2606.07082

Code (0)

등록된 구현이 없습니다.

Tasks

Reinforcement Learning

Similar Papers 제목 키워드 기반

GLaD: Geometric Latent Distillation for Vision-Language-Action Models

2025-12-10 · Minghao Guo, Meng Cao, Jiachen Tao, Rongtao Xu 외 arxiv

Most existing Vision-Language-Action (VLA) models rely primarily on RGB information, while ignoring geometric cues crucial for spatial reasoning and manipulation. In this work, we introduce GLaD, a geometry-aware VLA fra…

Knowledge DistillationSpatial Reasoning

Dense Supervision, Sparse Updates: On the Sparsity and Geometry of On-Policy Distillation

2026-06-11 · Guo Yu, Wenlin Liu, Yulan Hu, Hao-Xuan Ma 외 arxiv

On-policy distillation (OPD) has recently become a prominent post-training recipe by combining two desirable ingredients: on-policy student-generated trajectories and dense token-level teacher supervision. Yet how this h…

Visual-Advantage On-Policy Distillation for Vision-Language Models

2026-05-21 · Ruiqi Liu, Xiaolei Lv, Gengsheng Li, Ximo Zhu 외 arxiv

On-policy knowledge distillation has proven effective for language models, yet its application to vision-language models (VLMs) remains underexplored. We observe that standard on-policy distillation can improve a student…

Knowledge DistillationMathematical Reasoning

LIDEA: Human-to-Robot Imitation Learning via Implicit Feature Distillation and Explicit Geometry Alignment

2026-04-12 · Yifu Xu, Bokai Lin, Xinyu Zhan, Hongjie Fang 외 arxiv

Scaling up robot learning is hindered by the scarcity of robotic demonstrations, whereas human videos offer a vast, untapped source of interaction data. However, bridging the embodiment gap between human hands and robot …

PHF: Privileged Hidden Flow for On-Policy Self-Distillation

2026-06-28 · Yuhan Li, Mingxu Zhang, Dazhong Shen, Ying Sun arxiv

On-policy self-distillation (OPSD) trains a reasoning model on rollouts sampled from its own policy by matching a privileged teacher that also sees verified reference solutions. Existing OPSD objectives supervise only th…