发表机构
Beijing University of Posts and Telecommunications; University of Malaya; Georgian College; Huazhong University of Science and Technology（北京邮电大学; 马来亚大学; 乔治亚学院; 华中科技大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
JevSoup提出免训练LoRA组合框架，通过系统一路由选择两个专家并正交投影组合，在14个PorTAL任务和三个Qwen3规模上取得最高1.19%和1.21%的准确率提升。
AI 中文摘要
构建适应性AI系统需要跨不同任务有效协调专业化能力。低秩适应（LoRA）实现了模块化专长，但现有的路由方法可能需要辅助数据、额外训练或自回归解码。我们提出JevSoup，一个免训练框架，将系统一专家路由与系统二执行分离。仅使用输入和专家描述，Jev通过结构化概率选择两个专家。JevSoup保留领先专家的更新，将第二个专家投影到第一个更新行空间的正交补上，并以相等权重组合它们。在14个PorTAL任务和三个Qwen3规模上，JevSoup在任务宏和样本微准确率上相对于最强评估外部基线取得了高达1.19%和1.21%的绝对提升。我们的代码可在该https URL获取。
英文摘要
Building adaptable AI systems requires effective coordination of specialized capabilities across diverse tasks. Low-rank adaptation (LoRA) enables modular expertise, but existing routing approaches may require auxiliary data, additional training, or autoregressive decoding. We propose JevSoup, a training-free framework separating System One expert routing from System Two execution. Using only the input and expert descriptions, Jev selects two experts through structured probabilities. JevSoup retains the leading expert's update, projects the second onto the orthogonal complement of the first update's row space, and combines them with equal weights. Across 14 PorTAL tasks and three Qwen3 scales, JepSoup achieves absolute gains of up to 1.19\% in task-macro and 1.21\% in sample-micro accuracy over the strongest evaluated external baselines. Our code is available at https://github.com/Leowang980/JevSoup.
Comments5 pages, 2 figures, underreview
