This work proposes a recursive offline distillation method combining two components: entropy-based sampling for selective distillation and answer mixing, and a comparative analysis of methods to generate and normalize distillation traces.
Abstract
Distillation of reasoning chains from Large Language Models (LLMs) into their smaller siblings is a standard training approach. Frequently, researchers do not have full access to a teacher model and its internal representations, which limits the resulting performance. We propose a recursive offline distillation method combining two components: entropy-based sampling for selective distillation and answer mixing. Specifically, for three open models (3-4B), at each epoch, we sample reasoning chains weighted by entropy difference with an auxiliary medium model (70-72B) and, next, supplement thinking traces with single-token ground truth answers. Our approach consistently outperforms naive distillation with comparable data budgets ($57.15 \pm 1.01$ vs 51.61 on average) and offers a much steeper learning curve. Additionally, we present a comparative analysis of methods to generate and normalize distillation traces.
CaRE-KD is proposed, a confidence-gated distillation framework that replaces static objectives with uncertainty-adaptive optimization and provides a gradient-level analysis showing how this dual-granularity design induces a conditional calibration mechanism that prior static divergences cannot reproduce.
Small language models are often post-trained as students on reasoning traces from stronger teacher models to efficiently learn new skills. However, token-level imitation on traces that lie far outside the student's expected distribution often produces \textit{confident conflicts}, whereby the student is required to imi...
Juan Garcia Giraldo, Matteo Santelmo, E. Durech et al.· 0 citations
Autoregressive large language models (LLMs) have rapidly advanced in capability, but their increasing scale comes with substantial computational and memory costs at inference time. Knowledge distillation (KD) offers a practical solution by transferring knowledge from a large teacher model to a smaller student model via...
Byeonghu Na, Donghyeok Shin, Yeongmin Kim et al.· 0 citations
Explicit intermediate reasoning gives large language models (LLMs) a stronger problem-solving mode. We study learning from this think-mode advantage via on-policy distillation (OPD). OPD preserves student-generated trajectories and provides dense token-level teacher targets at student-visited prefixes. Privileged reaso...
Wan-Qi Ren, Jian-Xiang Wang, Dan-Xuan Liu et al.· 0 citations
On-policy self-distillation uses a model as its own teacher to provide dense supervision for reasoning, often through reference-solution conditioning. Providing privileged information does not by itself ensure effective token-level supervision throughout long responses. We introduce Activation-Conditioned Self-Distilla...
Zhe-Xi Lu, Subhajit Chaudhury, Tejaswini Pedapati et al.· 1 citation
On-policy self-distillation (OPSD) trains a large language model using supervision from a teacher instantiated from the same model as the student. The teacher is conditioned on privileged information (PI), such as verified solutions. The student, which does not observe PI, learns from the teacher's next-token distribut...
Jiazheng Li, Yawei Li, Jun-Ru Lu et al.· 0 citations
Learn how Google DeepMind and GovTech Singapore uses lightweight on-device Gemma models to intercept conversational scams in private chats with zero cloud data egress.
Microsoft Research Blog· microsoft.comOct 12, 2026
StudentSim is a research system for evaluating and improving AI tutoring tools before they are tested with learners. It models how individuals may engage with tutors, giving feedback to improve tutor performance across subjects like chess, writing, and mathematics. The post Simulating student learning for better AI tutoring appeared first on Microsoft Research.
Adaptive AI agents can help make BIM data more machine-readable by navigating IFC models, interpreting inconsistent information, and mapping it to defined standards. In this blog, Alok Rawat shares findings from a real-world pilot in construction workflows. The post Adaptive AI Agents in Construction Workflows appeared first on GPT-Lab.
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
