Adaptive Teacher Exposure for Self-Distillation in LLM Reasoning

Hugging Face Daily Papers Papers

Summary

Adaptive Teacher Exposure for Self-Distillation (ATESD) improves LLM reasoning by dynamically adjusting how much of the reference reasoning the teacher shows the student during training, using a learnable policy controller and a discounted learning-progress reward. Experiments on math benchmarks show consistent improvements over existing self-distillation and RL baselines.

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.
Original Article
View Cached Full Text

Cached at: 05/15/26, 08:24 AM

Paper page - Adaptive Teacher Exposure for Self-Distillation in LLM Reasoning

Source: https://huggingface.co/papers/2605.11458

Abstract

Adaptive Teacher Exposure for Self-Distillation (ATESD) improves large language model reasoning by dynamically adjusting teacher exposure during training through a learnable policy controller.

On-policyself-distillationhas 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-sideexposure 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 forSelf-Distillation(ATESD). ATESD models the reveal ratio with a lightweightBeta-policy controllerconditioned 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 adiscounted learning-progress rewardthat scores each held decision by its effect on the student’s future improvement rather than its immediate loss change, addressing thedelayed credit assignmentinduced by on-policy distillation. Experiments on AIME 24, AIME 25, and HMMT 25 across Qwen3-{1.7B, 4B, 8B} show that ATESD consistently outperforms competitiveself-distillationand 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 reasoningself-distillation.

View arXiv pageView PDFAdd to collection

Get this paper in your agent:

hf papers read 2605\.11458

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2605.11458 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2605.11458 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2605.11458 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Anti-Self-Distillation for Reasoning RL via Pointwise Mutual Information

Hugging Face Daily Papers

Proposes Anti-Self-Distillation (AntiSD) which reverses the knowledge transfer direction in self-distillation to improve math reasoning efficiency and accuracy, achieving GRPO baseline accuracy in 2-10x fewer steps and up to 11.5 points higher final accuracy across models from 4B to 30B parameters.

Improving Reasoning Capabilities in Small Models through Mixture-of-Layers Distillation with Stepwise Attention on Key Information

arXiv cs.CL

This paper proposes a novel Chain-of-Thought distillation framework that transfers teacher models' stepwise attention on key information to student models through a Mixture-of-Layers module for dynamic layer alignment. The method achieves consistent performance improvements on mathematical and commonsense reasoning benchmarks by explicitly guiding student models to progressively focus on critical information during reasoning.

Self-Distillation as a Performance Recovery Mechanism for LLMs: Counteracting Compression and Catastrophic Forgetting

arXiv cs.CL

This paper introduces Self-Distillation Fine-Tuning (SDFT) as a recovery mechanism for LLMs suffering from performance degradation due to catastrophic forgetting, quantization, and pruning. The authors provide theoretical justification using Centered Kernel Alignment (CKA) to demonstrate that self-distillation aligns the student model's high-dimensional manifold with the teacher's optimal structure, effectively recovering lost capabilities.