arXiv cs.AI· Yangjie Xu, Lujun Li, Lama Sleem, Niccolo Gentile, Yewei Song, Yiqun Wang, Siming Ji, Wenbo Wu, Hui Huang, Radu State·· 2 天前AI 评分32
Agent Skill 框架：中小型语言模型在工业环境中的潜力研究
Agent Skill Framework: Perspectives on the Potential of Small to Medium Language Models in Industrial Environments
AI 导读
一项大规模实证研究评估了 Agent Skill 范式在 270M–80B 开源模型上的可行性，覆盖保险理赔分类与 Python 编程两类任务。结果显示极小模型常无法检索或选择正确技能，而 30B–80B 模型从技能增强中获益显著。研究还给出面向部署的 GPU 显存时间占用（GB·min）画像，揭示技能增强收益随模型规模、任务复杂度与 GPU 预算的变化。
正文 · 原文
Abstract:Skills are getting increasingly used in recent agentic frameworks, performing robustly with proprietary LLMs. Yet, their effectiveness for small and medium-sized open-source models (270M--80B) remains underexplored, especially under industrial, resource-constrained conditions. As such, we present a large-scale empirical study evaluating the skill paradigm's feasibility across two open-source classification tasks: one insurance-claim task based on proprietary data, and a set of Python programming problems. We find that very small models frequently fail to retrieve or select correct skills, substantially limiting their usefulness in the evaluated skill-augmented settings, while larger 30B--80B models gain substantial benefits from skill augmentation. Furthermore, we provide a deployment-oriented GPU profile, reporting VRAM time occupancy (GB$\cdot$min) across models under our evaluation settings---a deployment-oriented metric that may help characterize GPU-memory costs, where small and medium models are typically preferred for VRAM control. These results characterize how the benefits of skill augmentation vary with model scale, task complexity, and GPU budget, and provide deployment-oriented observations.
| Comments: | 12 pages |
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2602.16653 [cs.AI] |
| (or arXiv:2602.16653v4 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2602.16653 arXiv-issued DOI via DataCite |
Submission history
From: Lujun Li [view email]
[v1]
Wed, 18 Feb 2026 17:52:17 UTC (78 KB)
[v2]
Fri, 20 Feb 2026 14:47:30 UTC (78 KB)
[v3]
Thu, 18 Jun 2026 19:19:49 UTC (820 KB)
[v4]
Thu, 15 Oct 2026 12:48:43 UTC (835 KB)
来源：arXiv cs.AI · arxiv.org