HuggingPapers· @HuggingPapers · X·· 12 小时前AI 评分
SoftwareGym 合成GUI环境训练智能体
SoftwareGym 一个新 pipeline 合成多样化的 GUI 软件环境来训练 computer-use 智能体,在 OSWorld 上实现 37% 的相对提升(Pass@3 从 41.4% 提升到 57.0%)
SoftwareGym
A new pipeline synthesizes diverse GUI software environments to train computer-use agents, achieving a 37% relative improvement on OSWorld (Pass@3 from 41.4% to 57.0%)
来源:HuggingPapers · x.com