Episode
On the Geometry of On-Policy Distillation
- Podcast
- Daily Paper Cast
- Published
- Jun 10, 2026
- Duration seconds
- 1613
- Processing state
not_requested- Canonical source
- https://share.transistor.fm/s/1b89b053
Actions
POST https://stenobird.com/v1/public/podcasts/daily-paper-cast-7079649/episodes/on-the-geometry-of-on-policy-distillation/transcription-requests
Idempotently request low-priority transcript generation for this episode.GET https://stenobird.com/podcast/daily-paper-cast-7079649/on-the-geometry-of-on-policy-distillation.md
Read the agent-friendly Markdown representation of this episode resource.
Summary
🤗 Upvotes: 59 | cs.LG, cs.AI Authors: Zhennan Shen, Yanshu Li, Qingyu Yin, Chak Tou Leong, Zhilin Wang, Yanxu Chen, Rongduo Han, Sunbowen Lee, Yi R. Fung Title: On the Geometry of On-Policy Distillation Arxiv: http://arxiv.org/abs/2606.07082v1 Abstract: 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.