paper-with-me

Papers

Why Does Self-Distillation (Sometimes) Degrade the Reasoning Capability of LLMs?

2026-03-25 · Jeonghye Kim, Xufang Luo, Minbeom Kim, Sangmook Lee, Dohyung Kim, Jiwon Jeon, Dongsheng Li, Yuqing Yang arxiv

Self-distillation has emerged as an effective post-training paradigm for LLMs, often improving performance while shortening reasoning traces. However, in mathematical reasoning, we find that it can reduce response length while degrading performance. We trace this degradation to the suppression of epistemic verbalization - the model's expression of uncertainty during reasoning. Through controlled experiments varying conditioning context richness and task coverage, we show that conditioning the teacher on rich information suppresses uncertainty expression, enabling rapid in-domain optimization with limited task coverage but harming OOD performance, where unseen problems benefit from expressing uncertainty and adjusting accordingly. Across Qwen3-1.7B/8B, DeepSeek-Distill-Qwen-7B, and Olmo3-7B-Instruct, we observe performance drops of up to 40%. Our findings highlight that exposing appropriate levels of uncertainty is crucial for robust reasoning and underscore the importance of optimizing reasoning behavior beyond merely reinforcing correct answer traces.

📄 PDF Abstract BibTeX arXiv:2603.24472

Code (0)

등록된 구현이 없습니다.

Tasks

Mathematical Reasoning

Similar Papers 제목 키워드 기반

Tailoring Teaching to Aptitude: Direction-Adaptive Self-Distillation for LLM Reasoning

2026-05-21 · Hongbin Zhang, Chaozheng Wang, Kehai Chen, Youcheng Pan 외 arxiv

On-policy self-distillation (OPSD) is an emerging LLM post-training paradigm in which the model serves as its own teacher: conditioned on privileged information such as a reference trace or hint, the same policy provides…

Mathematical Reasoning

OPSD Compresses What RLVR Teaches: A Post-RL Compaction Stage for Reasoning Models

2026-05-07 · Jaehoon Kim, Dongha Lee arxiv

On-Policy Self-Distillation (OPSD) has recently emerged as an alternative to Reinforcement Learning with Verifiable Rewards (RLVR), promising higher accuracy and shorter responses through token-level credit assignment fr…

Reinforcement LearningMathematical Reasoning

Training with Harnesses: On-Policy Harness Self-Distillation for Complex Reasoning

2026-05-09 · Zhengyang Zhao, Lu Ma, Wentao Zhang arxiv

Inference-time harnesses substantially improve large language models on complex reasoning tasks. However, the intrinsic capabilities of the underlying model remain unchanged by the addition of these external workflows. T…

Mathematical ReasoningText Classification

Self-Distilled Reasoner: On-Policy Self-Distillation for Large Language Models

2026-01-26 · Siyan Zhao, Zhihui Xie, Mengchen Liu, Jing Huang 외 arxiv

Knowledge distillation improves large language model (LLM) reasoning by compressing the knowledge of a teacher LLM to train smaller LLMs. On-policy distillation advances this approach by having the student sample its own…

Knowledge DistillationMathematical ReasoningReinforcement Learning

Rethinking On-Policy Self-Distillation for Thinking Models

2026-07-06 · Simran Kaur, Narutatsu Ri, Yinghui He, Liam Fowl 외 arxiv

Self-distillation is a promising recipe for self-improvement in language models. In this setting, a model can serve as its own teacher when given privileged information, such as a solution to a math problem. This seems e…