Adaptive Teacher Exposure for Self-Distillation in LLM Reasoning

Zihao Han, Tiangang Zhang, Huaibin Wang, Yilun Sun

arXiv:2605.11458 · 2026-05-15 공개 · arXiv · PDF

self-distillation llm-reasoning qwen3 on-policy aime hmmt learning-progress adaptive-exposure

Abstract

On-policy self-distillation has become a strong recipe for LLM reasoning, where a privileged teacher supervises the student's own rollouts while conditioning on the reference solution. A design choice shared by nearly all such methods, however, has gone unquestioned: the teacher always sees the full reference reasoning. We argue that this default itself is part of the problem and identify a teacher-side exposure mismatch: when the teacher conditions on reasoning far beyond the student's current competence, the resulting token targets become too strong to absorb. A controlled fixed-exposure sweep makes this concrete on two fronts: 1) full exposure is not reliably the best choice, and 2) student-teacher mismatch grows monotonically as the teacher sees more privileged reasoning. This motivates treating teacher exposure not as a fixed hyperparameter but as a learnable training-time control variable. We therefore propose Adaptive Teacher Exposure for Self-Distillation (ATESD). ATESD models the reveal ratio with a lightweight Beta-policy controller conditioned on compact training-state statistics, and uses one sampled exposure for a short hold window of student updates. To make this exposure controller learnable, we optimize it with a discounted learning-progress reward that scores each held decision by its effect on the student's future improvement rather than its immediate loss change, addressing the delayed credit assignment induced by on-policy distillation. Experiments on AIME 24, AIME 25, and HMMT 25 across Qwen3-{1.7B, 4B, 8B} show that ATESD consistently outperforms competitive self-distillation and RL baselines, improving over OPSD by +0.95, +2.05, and +2.33 Average@12 points respectively, and establishing adaptive teacher exposure as an effective new axis for reasoning self-distillation.

한국어 요약

📋 한 줄 요약

**[LLM 자기증류 / 적응적 교사 노출]** 자기증류에서 교사의 reference 노출 비율을 학습 가능한 제어 변수로 만들어 학습 진행에 따라 조정하는 ATESD 제안.

🎯 핵심 기여도

💡 핵심 아이디어

교사의 reference 노출은 단순한 hyperparameter가 아니라 학생의 현재 역량과 정합해야 하는 학습 시점 제어 변수이며, 즉각 손실이 아닌 학생의 미래 개선에 대한 기여로 평가해야 한다.

🔬 기술적 접근법

📊 주요 결과

💭 의의 및 한계

**의의**: 자기증류에서 교사 노출이라는 새로운 학습 차원을 정식화하고, 학생 역량에 맞춘 교사 정보 노출이 추론 학습의 핵심 lever임을 입증. **한계**: Beta-policy controller의 보상 함수 설계가 도메인에 민감, 매우 짧은 reasoning 태스크에서의 효과는 추가 검증 필요.

🚀 실용적 활용