arXiv cs.CL· Runsheng Chen, Yihong Liu, Raoyuan Zhao, Michael A. Hedderich, Hinrich Sch\"utze, Matthias Grabmair·· 13 小时前AI 评分31
Bridged Self-Distillation：统一框架实现跨分词器与闭源 API 教师的在线策略蒸馏
Bridged Self-Distillation for On-Policy Learning
AI 导读
研究者提出 Bridged Self-Distillation（BSD），一个可从任意提供推理轨迹的教师模型蒸馏的统一框架。外部教师仅需离线生成推理轨迹作为特权上下文，学生模型在无上下文条件下生成 rollout，再由条件化于该上下文的学生副本作为自教师，在学生自身 token 空间内为同一批 rollout 打分，并结合验证器提供的结果信号调整失败 rollout 的更新。
正文 · 原文
Abstract: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: this https URL
| Comments: | 21 pages, 10 figures, 15 tables. Code: this https URL |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.26397 [cs.CL] |
| (or arXiv:2610.26397v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.26397 arXiv-issued DOI via DataCite |
Submission history
From: Runsheng Chen [view email]
[v1]
Mon, 19 Oct 2026 13:40:50 UTC (1,029 KB)
[v2]
Tue, 27 Oct 2026 09:48:42 UTC (1,029 KB)
来源：arXiv cs.CL · arxiv.org