Runyang You, Yongqi Li, Tiezheng Yu, Wenjie Li · 2026 · arXiv (Cornell University)
DOI: 10.48550/arxiv.2610.28048 · cited by 0 works (OpenAlex) · open-access PDF
On-Policy Destillation (OPD) konsolidiert Expertenfähigkeiten in einen einzelnen Schüler, indem sie Lehrer- und Schülerabschnittsverteilungen auf Studentengenerierten Rollouts miteinander verknüpft. Wir untersuchen, ob die Tiefenvorhersage-Übergangsformen, die Änderungen der Prognosen zwischen den Schichten, zusätzliche Aufgabenrelevante Überwachung bieten können. Unsere Diagnosestudie stellt fest, dass eine stärkere Lehrer-Studentenvereinbarung bei Zwischenübergangsformen mit einer höheren Antwortkorrektheit innerhalb von Fragen verbunden ist, wobei korrekte Antworten einen niedrigeren Durchschnittswert von KL über alle gemessenen Zwischenschichtenpaare zeigen. Diese Beobachtungen motivieren On-Policy Transition Destillation (OPTD), die die
Machine translation (Meta M2M100); the English original follows.
On-policy distillation (OPD) consolidates expert capabilities into a single student by matching teacher and student final-layer distributions on student-generated rollouts. We investigate whether depthwise prediction transitions, the changes in predictions between layers, can provide additional task-relevant supervision. Our diagnostic study finds that stronger teacher--student agreement in intermediate transitions is associated with higher answer correctness within questions, with correct responses exhibiting lower mean transition KL across all measured intermediate layer pairs. These observations motivate On-Policy Transition Distillation (OPTD), which augments final-layer distribution matching by aligning teacher and student layer-to-layer log-probability increments on student-generated rollouts. Evaluations across three backbones and eight benchmarks spanning mathematics, coding, and instruction following show that OPTD achieves the best or tied-best overall performance in both dual- and three-domain distillation. In the three-domain setting, OPTD recovers $\approx 90\%$ of the overall teacher--student capability gap, compared with $\approx 75\%$ for OPD.
Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese
Topics: Distillation, Matching (statistics), Correctness, Transition (genetics), Computer science, Distribution (mathematics)
GET https://store.agentexchange.work/gw/papers/verify?title=...&year=... (0.003 USD), fetch this record as JSON with /gw/papers/paper?id=2610.28048, or list what it cites with /gw/papers/references?id=.... Pay per call in USDC (x402) or with a 19 USD card key for 1,000 calls. Catalogue · llms.txtSources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange