arXiv cs.AI· Ik-hwan Kim, Hyeongrok Han, Mingi Jung, Sangwon Yu, Jinseok Hong, Sang Hun Kim, Yoonyoung Choi, Sungroh Yoon·· 13 小时前AI 评分31
通过智能体编辑蒸馏元认知行为：MBT-A 让 Qwen3.5-4B-Base 数学得分达 85.4
Distilling Metacognitive Behaviors through Agentic Editing
AI 导读
研究者提出元认知行为调优（MBT），在教师推理轨迹缺失处插入七种元认知行为后再训练学生模型；其智能体方法 MBT-A 逐行编辑轨迹并以 git commit 记录每次修改，避免长轨迹被摘要化、插入点过度集中。在竞赛数学上，用 MBT-A 轨迹训练的 Qwen3.5-4B-Base 学生得分 85.4，高于未编辑轨迹 SFT 的 83.1，并优于 token 匹配对照和五种先前方法。
正文 · 原文
Abstract:Distilling long reasoning traces from a teacher model into a smaller student also passes on the teacher's reasoning habits. Prior work and our annotation of teacher traces show that current reasoning models often start without a plan, verify superficially, abandon promising solution paths, and take unchecked shortcuts. We propose Metacognitive Behavioral Tuning (MBT), which inserts seven metacognitive behaviors into teacher traces at the points where the reasoning lacks them and then trains the student on the edited traces. Because teacher traces are long, how the edits are made matters. Rewriting a whole trace at once summarizes long traces, and choosing all insertion points at once crowds them at the start and end. Our agentic method, MBT-A, edits the trace file line by line and records each edit as a git commit, which avoids summarizing long traces and spreads the insertions. On competition math, a Qwen3.5-4B-Base student trained on MBT-A traces scores 85.4 against 83.1 for SFT on the unedited traces and also outperforms a token-matched control and five prior methods. MBT-A improves two other students as well, and the inserted behaviors clearly transfer to the students' reasoning, with rare ones several times more frequent.
| Comments: | 25 pages. Code and data: this https URL |
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2602.22508 [cs.AI] |
| (or arXiv:2602.22508v3 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2602.22508 arXiv-issued DOI via DataCite |
Submission history
From: Ik-Hwan Kim [view email]
[v1]
Thu, 26 Feb 2026 00:56:15 UTC (1,283 KB)
[v2]
Mon, 11 May 2026 05:22:14 UTC (1,232 KB)
[v3]
Tue, 27 Oct 2026 07:39:06 UTC (456 KB)
来源：arXiv cs.AI · arxiv.org