Episode
Trust Region On-Policy Distillation
- Podcast
- Daily Paper Cast
- Published
- Jun 4, 2026
- Duration seconds
- 1445
- Processing state
not_requested- Canonical source
- https://share.transistor.fm/s/2bb037b7
Actions
POST https://stenobird.com/v1/public/podcasts/daily-paper-cast-7079649/episodes/trust-region-on-policy-distillation/transcription-requests
Idempotently request low-priority transcript generation for this episode.GET https://stenobird.com/podcast/daily-paper-cast-7079649/trust-region-on-policy-distillation.md
Read the agent-friendly Markdown representation of this episode resource.
Summary
🤗 Upvotes: 33 | cs.LG, cs.CL Authors: Xingrun Xing, Haoqing Wang, Boyan Gao, Ziheng Li, Yehui Tang Title: Trust Region On-Policy Distillation Arxiv: http://arxiv.org/abs/2606.01249v2 Abstract: On-Policy Distillation (OPD) is a fundamental technique for efficient post-training of large language models (LLMs), with broad applications in agent learning, multi-task enhancement, and model compression. However, OPD training becomes unstable when the teacher and student distributions differ substantially, as teacher supervision on student-generated tokens may yield unreliable policy gradients and even cause optimization failure. This work addresses reliable on-policy token-level supervision through credit assignment strategies, and proposes Trust Region On-Policy Distillation, TrOPD. It features the following characteristics: 1) Trust-Region On-Policy Learning: TrOPD performs OPD only in regions where the teacher provides reliable supervision, mitigating the optimization difficulty of the K1 reverse-KL estimator under distribution mismatch. 2) Outlier Estimation: For outlier regions, we explore gradient clipping, masking, and forward-KL estimation to reduce the adverse effects of unreliable supervision. 3) Off-Policy Guidance: The student continues generation from teacher prefixes and uses forward KL to imitate off-policy guidance, encouraging on-policy exploration toward reliable regions. Experiments show that TrOPD consistently outperforms SoTA OPD baselines, including OPD, EOPD, and REOPOLD, across mathematical reasoning, code generation, and general-domain benchmarks.