DAPD：双锚定策略蒸馏缓解语言模型特权幻觉

AI 摘要

本文提出了一种名为DAPD（双锚定策略蒸馏）的新框架，用于解决语言模型后训练中在线策略自蒸馏（OPSD）存在的特权幻觉问题。DAPD通过双路径锚定和双源锚定两个层面的锚定机制，消除了教师模型与学生模型在推理时的信息不对称。实验表明，DAPD在Qwen3-4B上平均提升2.00分，在4B和32B规模上分别提升2.69和2.78分，显著优于OPSD。

正文节选

DAPD: Dual-Anchored Policy Distillation Abstract On-policy (self) distillation (OPSD) is increasingly adopted for language-model post-training. It strengthens the teacher with privileged information but can induce a privilege illusion: the student learns privilege-dependent behavior it cannot reproduce from its inference-time context, yet behaves as if the training-time privileged information remained available, ultimately degrading performance. In this paper, we identify information asymmetry b

发布时间：—

抓取时间：2026-08-04 14:13

来源机构：Hugging Face