# Adaptive Teacher Exposure for Self-Distillation in LLM Reasoning

Source: [Adaptive Teacher Exposure for Self-Distillation in LLM Reasoning](<https://www.sota2.com/research/paper/6a0405badbc836b87d8276de>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-06-11T05:00:05.512Z

Authors: Zihao Han, Tiangang Zhang, Huaibin Wang, Yilun Sun

Published: 2026-05-12T03:15:58.000Z

## 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.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Mathematical Reasoning on AIME 2024 (Average@12 accuracy)](<https://www.sota2.com/research/sota/mathematical-reasoning-on-aime-2024-average-at-12-accuracy.md>): Mathematical Reasoning; AIME 2024; Average@12 Accuracy (AIME 2024): 80.56.

- [Mathematical Reasoning on AIME 2025 (Average@12 accuracy)](<https://www.sota2.com/research/sota/mathematical-reasoning-on-aime-2025-average-at-12-accuracy.md>): Mathematical Reasoning; AIME 2025; Average@12 Accuracy: 72.5.

- [Mathematical Reasoning on HMMT 2025](<https://www.sota2.com/research/sota/mathematical-reasoning-on-hmmt-2025.md>): Mathematical Reasoning; HMMT 2025; Accuracy: not recorded.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2605.11458>)

[PDF](<https://arxiv.org/pdf/2605.11458.pdf>)
