Spreadsheet-RL: Advancing Large Language Model Agents on Realistic Spreadsheet Tasks via Reinforcement Learning
Spreadsheet-RL: 通过强化学习推进大型语言模型代理在现实中的电子表格任务中的进步
机构 * University of Illinois Urbana-Champaign(伊利诺伊大学厄巴纳-香槟分校) ; Meta
AI总结 本文提出Spreadsheet-RL,一种通过强化学习微调框架,旨在在现实Microsoft Excel环境中训练专门的电子表格代理。该方法通过自动化管道收集在线论坛中的配对起始-目标电子表格,以及金融和供应链管理等领域的领域特定评估任务,构建了新的Domain-Spreadsheet基准数据集,并展示了在通用和领域特定电子表格任务上的显著性能提升。
Comments Mingyuan served as the project lead. Banghao, Yining, and Mingyuan contributed equally to this work, with more junior authors listed before senior authors. All data and code releases are maintained by the corresponding authors at UIUC and are not affiliated with Meta