2605.28791v1 May 27, 2026 cs.CL

Skill-Conditioned Gated Self-Distillation for LLM Reasoning

Senkang Hu
Senkang Hu
Citations: 499
h-index: 12
Yong Dai
Yong Dai
Citations: 31
h-index: 3
Yuzhi Zhao
Yuzhi Zhao
Citations: 95
h-index: 6
Xiao Luo
Xiao Luo
Citations: 51
h-index: 4
Jiazhe Huang
Jiazhe Huang
Citations: 0
h-index: 0
Xiao Chen
Xiao Chen
Citations: 17
h-index: 2

On-policy self-distillation (SD) improves LLM reasoning by using teacher-side privileged information (PI) to turn sparse verifier outcomes into dense token-level supervision. Existing methods usually assume trusted PI, such as reference answers or successful traces. We ask whether PI can instead come from an experience-derived skill bank, where retrieved skills are compact and reusable but may also be irrelevant or misleading. We propose Skill-Conditioned Gated Self-Distillation (SGSD), which formulates skill-based SD as teacher hypothesis validation rather than unconditional imitation. SGSD retrieves skill-mistake pairs, constructs a multi-teacher pool, and lets all skill-conditioned teachers score the same plain-prompt student rollout. The verifier validates each teacher's polarity: supporting a success or suppressing a failure gives positive supervision, while the opposite stance is reversed. A robust gated objective then distills informative teacher-student disagreements while suppressing uncertain or extreme signals. Experiments on multiple mathematical reasoning benchmarks show that SGSD consistently improves over GRPO and remains competitive with answer-conditioned OPSD under a weaker PI assumption. For example, on Qwen3-1.7B, SGSD outperforms GRPO by 6.2% and OPSD by 1.7% on average on AIME24, AIME25, and HMMT25. Our code is available at https://github.com/walawalagoose/SGSD.

0 Citations
0 Influential
32.931471805599 Altmetric
164.7 Score
Original PDF
3

No Analysis Report Yet

This paper hasn't been analyzed by Gemini yet.

Log in to request an AI analysis.

댓글

댓글을 작성하려면 로그인하세요.

아직 댓글이 없습니다. 첫 번째 댓글을 남겨보세요!