arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2610.17670cs.AI
收藏

Cowork-Gym:规模化合成交互式工作环境

Cowork-Gym: Synthesizing Interactive Work Environments at Scale

Alibaba Token Foundry(阿里巴巴令牌工厂)
Anni Zou, Cheng Chen, Jiaji Deng, Weikang Zhou, Xinji Mai, Yue Wu, Jianing Lin, Lipeng Xie, Lixiong Qin, Qingyue Zhao, Que Shen, Shuchang Tao, Xuchen Pan, Yalia… 展开作者

Anni Zou, Cheng Chen, Jiaji Deng, Weikang Zhou, Xinji Mai, Yue Wu, Jianing Lin, Lipeng Xie, Lixiong Qin, Qingyue Zhao, Que Shen, Shuchang Tao, Xuchen Pan, Yaliang Li, Yunpeng Zhai, Yushuo Chen, Zhihai Wang, Zhuo Zhang, Zhaoyang Liu, Bolin Ding, Jingren Zhou

AI总结:

本研究推出Cowork-Gym框架,合成6.49万个工作环境后对不同规模Qwen模型进行后训练,相关模型在智能体基准测试中表现优于基础模型,其中Ours-27B在两个基准测试中超越了参数超100B的模型。

摘要:

Cowork智能体在工具和文件间处理依赖关系,同时适配不断变化的用户需求及对共享工作空间的编辑。规模化此类环境需维持任务目标、可执行资源、用户行为与成功标准间的一致性。我们推出Cowork-Gym框架,该框架通过构建-执行-修正循环合成Cowork环境,此循环利用执行证据与审查反馈验证并修复生成的环境。为捕捉真实的Cowork动态,一个有状态用户模拟器通过对话及对共享工作空间的直接编辑响应智能体的动作。我们在14个应用场景中合成了64900个环境,并用它们对9B、27B和35B-A3B的Qwen模型进行后训练,采用监督微调、强化学习和on-policy蒸馏方法。在6个智能体基准测试中,这些模型的平均表现较各自的基础检查点分别提升了8.7、6.4和4.1个百分点。Ours-27B在所有评估模型中排名第二,且在PawBench和ClawEval上取得最高平均分数,在这两个基准测试中超越了总参数规模超100B的模型。

英文摘要:

Cowork agents navigate dependencies across tools and files while adapting to changing user requirements and edits to shared workspaces. Scaling such environments requires maintaining consistency among task objectives, executable resources, user behavior, and success criteria. We introduce Cowork-Gym, a framework that synthesizes Cowork environments through a build-execute-revise loop that uses execution evidence and review feedback to validate and repair generated environments. To capture realistic Cowork dynamics, a stateful user simulator responds to the agent's actions through dialogue and direct edits to the shared workspace. We synthesize 64.9K environments across 14 application scenarios and use them to post-train 9B, 27B, and 35B-A3B Qwen models with supervised fine-tuning, reinforcement learning, and on-policy distillation. Averaged across six agentic benchmarks, these models outperform their corresponding base checkpoints by 8.7, 6.4, and 4.1 points, respectively. Ours-27B ranks second overall among the evaluated models and achieves the highest average scores on PawBench and ClawEval, outperforming models with over 100B total parameters on both benchmarks.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑