---
license: apache-2.0
language:
- en
base_model:
- Qwen/Qwen2.5-32B-Instruct
---

# AgentTrek: Agent Trajectory Synthesis via Guiding Replay with Web Tutorials

<div align="center">

[\[🏠Homepage\]](https://agenttrek.github.io/) [\[💻Code\]](https://github.com/xlang-ai/AgentTrek) [\[📝Paper\]](https://arxiv.org/abs/2412.09605) [\[🤗Models\]](https://huggingface.co/xlangai/AgentTrek-1.0-32B)[\[🤗Data\]]()  

</div>

## Overview of Pipeline

![image/png](https://cdn-uploads.huggingface.co/production/uploads/669ca7e678115e16bdfc9bfc/ULRzaG6fLCdR7ThkPyKwg.png)

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](https://huggingface.co/Qwen/Qwen2.5-32B-Instruct). 

![image/png](https://cdn-uploads.huggingface.co/production/uploads/669ca7e678115e16bdfc9bfc/ET8fq3iG-1N-lmy_ltNuH.png)

- For metrics, refers to [Browsergym Leaderboard](https://huggingface.co/spaces/ServiceNow/browsergym-leaderboard)
- For evaluation, refers to [Evaluation Script](https://github.com/xlang-ai/AgentTrek)
- For training dataset, refers to [Training Dataset]()

## Citation
```bibtex
@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}
}
```