pipette
ENEnglish

Bridged Self-Distillation for On-Policy Learning

Runsheng Chen, Yihong Liu, Raoyuan Zhao, Michael A. Hedderich, Hinrich Sch\"utze, Matthias Grabmair Technical University of Munich, Germany, LMU Munich, Germany, Munich Center for Machine Learning, Germany)

PreprintCódigo disponible

En palabras de los autores

On-policy distillation can transfer the reasoning capabilities of a strong teacher to a smaller student. However, distillation from a stronger teacher model faces two mismatches: its solutions may be difficult for a smaller student to imitate, and its token-level probabilities may be incompatible with, or even unavailable to, the student. Existing methods handle these settings through different forms of supervision. To this end, we propose Bridged Self-Distillation (BSD), a unified framework for distilling from any teacher that provides reasoning traces. The model to be distilled from, termed the external teacher, first provides a reasoning trace as privileged context, which can be generated in an offline fashion. The student generates rollouts without this context, while a copy of the student conditioned on it, serving as the self-teacher, scores the same rollouts in the student's own token space. BSD uses the resulting token-level guidance to adjust updates on unsuccessful rollouts, with a verifier providing the outcome signal. As the external teacher is queried only to produce text, the same training procedure applies to all teacher scenarios: shared-tokenizer, cross-tokenizer, and closed API teachers. Across these scenarios, BSD improves over the base model and outperforms baselines tailored to each scenario, offering a novel and unified pathway for distillation. Code is available: https://github.com/TUMLegalTech/Bridged-Self-Distillation

Resultado principalEl resumen no menciona limitaciones.

Apareció: martes, 20 de octubre. arXiv. Preprint, todavía sin revisión por pares.

Comentario de los autores: 21 pages, 10 figures, 15 tables. Code: https://github.com/TUMLegalTech/Bridged-Self-Distillation