AgentTrek: Agent Trajectory Synthesis via Guiding Replay with Web Tutorials

[🏠Homepage] [💻Code] [📝Paper] [🤗Models][🤗Data]

Overview of Pipeline

AgentTrek is a cost-efficient and scalable framework that synthesizes high-quality agent trajectories by guiding replay with web tutorials. These collected trajectories significantly enhance agent performance.

Quick Start

AgentTrek-1.0-32B is a web agent model finetuned from Qwen2.5-32B-Instruct.

• For metrics, refers to Browsergym Leaderboard
• For evaluation, refers to Evaluation Script
• For training dataset, refers to Training Dataset

Citation

@article{xu2024agenttrek,
  author    = {Yiheng Xu and Dunjie Lu and Zhennan Shen and Junli Wang and Zekun Wang and Yuchen Mao and Caiming Xiong and Tao Yu},
  title     = {AgentTrek: Agent Trajectory Synthesis via Guiding Replay with Web Tutorials},
  year={2024},
  eprint={2412.09605},
  archivePrefix={arXiv},
  primaryClass={cs.CL},
  url={https://arxiv.org/abs/2412.09605}
}