Verify Before You Distill: Prompt-Level Teacher Gating for On-Policy Distillation</p>\n","updatedAt":"2026-09-08T02:29:54.592Z","author":{"_id":"65328aa39326d6da5ff19b52","avatarUrl":"/avatars/5c3de984cd6eba69616bb608796865c5.svg","fullname":"Fei Zhao","name":"Hiiamein","type":"user","isPro":false,"isHf":false,"isHfAdmin":false,"isMod":false,"followerCount":4,"isUserFollowing":false}},"numEdits":0,"identifiedLanguage":{"language":"en","probability":0.3938365876674652},"editors":["Hiiamein"],"editorAvatarUrls":["/avatars/5c3de984cd6eba69616bb608796865c5.svg"],"reactions":[],"isReport":false}}],"primaryEmailConfirmed":false,"paper":{"id":"2609.02998","authors":[{"_id":"6a9a29e68f7c3b755723945a","name":"Zhiwei Zhang","hidden":false},{"_id":"6a9a29e68f7c3b755723945b","name":"Zechen Sun","hidden":false},{"_id":"6a9a29e68f7c3b755723945c","user":{"_id":"65328aa39326d6da5ff19b52","avatarUrl":"/avatars/5c3de984cd6eba69616bb608796865c5.svg","isPro":false,"fullname":"Fei Zhao","user":"Hiiamein","type":"user","name":"Hiiamein"},"name":"Fei Zhao","status":"claimed_verified","statusLastChangedAt":"2026-09-08T07:37:51.949Z","hidden":false},{"_id":"6a9a29e68f7c3b755723945d","name":"Kang Peng","hidden":false},{"_id":"6a9a29e68f7c3b755723945e","name":"Bin Liang","hidden":false},{"_id":"6a9a29e68f7c3b755723945f","name":"Huayu Deng","hidden":false},{"_id":"6a9a29e68f7c3b7557239460","name":"Yao Hu","hidden":false},{"_id":"6a9a29e68f7c3b7557239461","name":"Kam-Fai Wong","hidden":false},{"_id":"6a9a29e68f7c3b7557239462","name":"Mu Chuan","hidden":false}],"publishedAt":"2026-09-02T00:00:00.000Z","submittedOnDailyAt":"2026-09-08T00:00:00.000Z","title":"Verify Before You Distill: Prompt-Level Teacher Gating for On-Policy Distillation","submittedOnDailyBy":{"_id":"65328aa39326d6da5ff19b52","avatarUrl":"/avatars/5c3de984cd6eba69616bb608796865c5.svg","isPro":false,"fullname":"Fei Zhao","user":"Hiiamein","type":"user","name":"Hiiamein"},"summary":"On-policy distillation (OPD) accelerates post-training by providing dense token-level supervision from a frozen teacher on the student's own rollouts. Vanilla OPD applies this supervision uniformly across prompts, without checking whether the teacher is reliable for each prompt. Because reverse KL is mode-seeking, a confidently wrong teacher can induce a strong yet misleading update. Distributional proxies, such as entropy or teacher-student likelihood agreement, measure uncertainty or agreement but do not directly verify outcome correctness. We introduce Teacher-Gated On-Policy Distillation (TGOPD), built on the principle that teacher reliability should be verified at the prompt level before dense supervision is admitted. TGOPD estimates reliability from a small set of verifier-scored teacher probes and routes each prompt exclusively to dense OPD when the reliability check passes or to verifier-grounded GRPO otherwise. Across 4B and 35B students in mathematics, code, and instruction following, TGOPD outperforms Vanilla OPD in all six single-domain settings and achieves higher seven-benchmark averages at both scales under multi-domain training. By using otherwise-idle teacher capacity for reliability estimation, TGOPD also reduces teacher-side compute waste in asynchronous OPD, increasing teacher-node GPU utilization from 9.8% to 78.9% in the measured 4B single-domain run.","upvotes":5,"discussionId":"6a9a29e68f7c3b7557239463","ai_summary":"Teacher-Gated On-Policy Distillation verifies teacher reliability per prompt via verifier-scored probes, routing to dense distillation or verifier-grounded reinforcement learning to improve training efficiency and accuracy.","ai_keywords":["on-policy distillation","reverse KL","teacher reliability","verifier-scored probes","TGOPD","GRPO","asynchronous OPD"],"ai_summary_model":"thinkingmachines/Inkling-Small"},"canReadDatabase":false,"canManagePapers":false,"canSubmit":false,"hasHfLevelAccess":false,"upvoted":false,"upvoters":[{"_id":"65328aa39326d6da5ff19b52","avatarUrl":"/avatars/5c3de984cd6eba69616bb608796865c5.svg","isPro":false,"fullname":"Fei Zhao","user":"Hiiamein","type":"user"},{"_id":"678b35ef0d44299225201c71","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/O8DWzE62cSFclPXpoE_vT.png","isPro":false,"fullname":"zhangzhiwei","user":"zhangzhiwei666","type":"user"},{"_id":"68b2a4157f881fc640ba7d80","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/lMTgr3pe7pOHtMe7bVF7F.png","isPro":false,"fullname":"khtsly","user":"khtsly","type":"user"},{"_id":"68beaccde52df040b2d85b30","avatarUrl":"/avatars/9bfffe1d1290e920cf7efefc396f4019.svg","isPro":false,"fullname":"licanwei","user":"Powder021","type":"user"},{"_id":"63ac5701c21e60a3e9b58aa7","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/63ac5701c21e60a3e9b58aa7/g6EX7diOpuA94R2ab-rZC.png","isPro":true,"fullname":"Dipankar Sarkar","user":"dipankarsarkar","type":"user"}],"acceptLanguages":["en"],"dailyPaperRank":0,"markdownContentUrl":"https://huggingface.co/buckets/huggingchat/papers-content/resolve/2609/2609.02998.md","query":{}}">

Get this paper in your agent:

Don't have the latest CLI?

Cite arxiv.org/abs/2609.02998 in a model README.md to link it from this page.

Cite arxiv.org/abs/2609.02998 in a dataset README.md to link it from this page.

Cite arxiv.org/abs/2609.02998 in a Space README.md to link it from this page.

Add this paper to a collection to link it from this page.