arXiv cs.CL· Duy Nguyen, Hanqi Xiao, Archiki Prasad, Zaid Khan, Anirban Das, Shi-Xiong Zhang, Sambit Sahu, Hyunji Lee, Elias Stengel-Eskin, Mohit Bansal·· 10 小时前AI 评分27
AVSD:通过平衡共识与教师专属特权信号实现自适应视角自蒸馏
AVSD: Adaptive-View Self-Distillation by Balancing Consensus and Teacher-Specific Privileged Signals
AI 导读
AVSD 提出一种多特权信息视角的自蒸馏方法,将跨视角稳定的共识信号与视角专属残差信号分离,仅在残差与共识方向一致且比例适当时才用于调整更新幅度。在 AIME24、AIME25、HMMT25 上,AVSD 在 Qwen3-8B 和 Qwen3-4B 上分别比最强基线平均提升 Avg@8 3.1% 和 2.2%,并优于 GRPO。
正文 · 原文
Abstract:Self-distillation enables language models to learn on-policy from their own trajectories by using the same model as both student and teacher, with the teacher being conditioned on privileged information unavailable to the student. Such information can come in different types or views, such as solutions, demonstrations, feedback, or final answers. This setup provides dense token-level feedback without relying on a separate external model, but creates a fundamental asymmetry: the teacher may rely on view-specific information that the student cannot access at inference time. Moreover, the best type of privileged information is often task-dependent, making it difficult to choose a single teacher view. In this work, we address both these challenges jointly by introducing AVSD (Adaptive-View Self-Distillation), a novel method of self-distillation with multiple privileged-information views, which reconstructs token-level supervision by separating stable cross-view consensus from view-specific residual signals. AVSD identifies the consensus signal shared across views, which provides a reliable update direction, and then selectively adds the view-specific residual signal to adjust the update magnitude when it both aligns with the consensus direction and remains proportionate to the consensus signal. Experiments on math competition benchmarks (AIME24, AIME25, and HMMT25) show that AVSD consistently outperforms both single-view self-distillation baselines and GRPO, achieving average Avg@8 gains of 3.1% and 2.2% over the strongest baselines on Qwen3-8B and Qwen3-4B, respectively. Moreover, on code-generation benchmarks (Codeforces, LiveCodeBench v6) using Qwen3-8B, AVSD outperforms the single-view self-distillation baseline by 2.4% on average.
| Comments: | Accepted to NeurIPS 2026 |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Computation and Language (cs.CL) |
| Cite as: | arXiv:2605.20643 [cs.LG] |
| (or arXiv:2605.20643v2 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2605.20643 arXiv-issued DOI via DataCite |
Submission history
From: Duy Nguyen [view email]
[v1]
Wed, 20 May 2026 03:06:36 UTC (1,988 KB)
[v2]
Thu, 19 Nov 2026 01:34:42 UTC (1,990 KB)
来源:arXiv cs.CL · arxiv.org