Skill-Conditioned Gated Self-Distillation for LLM Reasoning
For practitioners of LLM reasoning, SGSD improves self-distillation under weaker privileged information assumptions, offering a practical alternative when reference answers are unavailable.
The paper proposes Skill-Conditioned Gated Self-Distillation (SGSD) for LLM reasoning, which uses retrieved skill-mistake pairs as privileged information instead of trusted reference answers. On Qwen3-1.7B, SGSD outperforms GRPO by 6.2% and OPSD by 1.7% on average across AIME24, AIME25, and HMMT25 benchmarks.
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.