MixSD: Mixed Contextual Self-Distillation for Knowledge Injection

Jiarui Liu, Lechen Zhang, Yongjin Yang, Yinghui He, Yingheng Wang, Weihao Xuan, Zhijing Jin, Mona Diab

arXiv:2605.16865 · 2026-05-19 공개 · arXiv · PDF

language-models self-distillation supervised-fine-tuning catastrophic-forgetting factual-recall model-capability fisher-sensitive knowledge-injection

Abstract

Supervised fine-tuning (SFT) is widely used to inject new knowledge into language models, but it often degrades pretrained capabilities such as reasoning and general-domain performance. We argue this forgetting arises because fine-tuning targets from humans or external systems diverge from the model's autoregressive distribution, forcing the optimizer to imitate low-probability token sequences. To address this problem, we propose MixSD, a simple external-teacher-free method for distribution-aligned knowledge injection. Instead of training on fixed targets, MixSD constructs supervision dynamically by mixing tokens from two conditionals of the base model itself: an expert conditional that observes the injected fact in context, and a naive conditional that reflects the model's original prior. The resulting supervision sequences preserve the factual learning signal while remaining substantially closer to the base model's distribution. We evaluate MixSD on two synthetic corpora that we construct to study factual recall and arithmetic function acquisition in a controlled setting, together with established benchmarks for open-domain factual question answering and knowledge editing. Across multiple model scales and settings, MixSD consistently achieves a better memorization-retention trade-off compared to SFT and on-policy self distillation baselines, retaining up to 100% of the base model's held-out capability while maintaining near-perfect training accuracy, whereas standard SFT retains as little as 1%. We further show that MixSD produces substantially lower-NLL supervision targets under the base model and reduces harmful movement along Fisher-sensitive parameter directions. These results suggest that aligning supervision with the model's native generation distribution is a simple and effective principle for knowledge injection that mitigates catastrophic forgetting.

한국어 요약

📋 한 줄 요약

**[지식 주입 / 자기증류]** 모델의 expert/naive conditional을 혼합해 분포 정합적 감독 신호를 만드는 외부 교사 없는 지식 주입 방법 MixSD 제안.

🎯 핵심 기여도

💡 핵심 아이디어

지식 주입 후 망각의 본질은 "target이 모델 분포와 멀다는 것"이며, base 모델 자신을 두 conditional로 분리해 감독 신호를 합성하면 분포 정합성을 자연스럽게 회복할 수 있다.

🔬 기술적 접근법

📊 주요 결과

💭 의의 및 한계

**의의**: catastrophic forgetting 완화에 외부 교사가 필요 없다는 강한 주장과 함께 분포 정합 감독 신호 설계의 원리를 제공. **한계**: expert/naive 두 conditional을 모두 base 모델 자신으로 얻어야 하므로 매우 짧은 컨텍스트로 충분히 표현되지 않는 지식 형태에서는 검증 필요.

🚀 실용적 활용