arXiv · Artificial Intelligence· Sida Li, Yuanxin Liu, Shicheng Li, Linli Yao, Xu Sun, Haoli Bai, Xiaohui Li·· 4 hr agoAI score52
TQS:用轨迹结构评分筛选编码智能体 SFT 数据
Quality, Not Just Outcome: A Structural Scoring Framework for Selecting Coding-Agent Trajectories for Supervised Fine-Tuning
AI brief
论文提出 TQS(Trajectory Quality Score),一个直接基于轨迹结构、无需 LLM judge 且线性时间运行的编码智能体轨迹评分框架。TQS 围绕信息聚焦、意图—动作对齐和策略动态刻画轨迹质量,在 155,144 条公开许可语料的成功轨迹上按 TQS 排序并微调,SWE-Bench-Verified 上高分轨迹训练结果优于随机或低分子集。
Source: arXiv · Artificial Intelligence · arxiv.org