26hzhang / os-genesis

Code and data for OS-Genesis: Automating GUI Agent Trajectory Construction via Reverse Task Synthesis

OS-Genesis

This repository contains the code and data for paper OS-Genesis: Automating GUI Agent Trajectory Construction via Reverse Task Synthesis.

We are uploading the data and checkpoints. Due to bandwidth limitations, this will take some time. Stay tuned!

Overview

We introduce OS-Genesis, an interaction-driven pipeline for synthesizing high-quality and diverse GUI agent trajectory data without human supervision or predefined tasks. By leveraging reverse task synthesis and a trajectory reward model, OS-Genesis enables effective end2end training of GUI agents.

Training

For details and operations of the training, please refer to the InternVL2 documentation and Qwen2-VL.

Evaluation

AndroidControl

To evaluate the AndroidControl Benchmark, please follow the steps below:

1. Clone the GitHub Repository:

   git clone https://github.com/OS-Copilot/OS-Genesis.git
   

2. Inference:

   cd OS-Genesis/evaluation
   bash run_ac_inference.sh
   

3. Evaluation:

   pyhton ac_eval.py
   

Mobile

Model Name	Base Model	Training Data	HF Link
OS-Genesis-4B-AC	InternVL2-4B	OS-Genesis-ac-training-data	🤗 link
OS-Genesis-7B-AC	Qwen2-VL-7B-Instruct	OS-Genesis-ac-training-data	🤗 link
OS-Genesis-8B-AC	InternVL2-8B	OS-Genesis-ac-training-data	🤗 link

Web

(Coming Soon)

Citation 📖

🫶 If you are interested in our work or find this repository / our data helpful, please consider using the following citation format when referencing our paper:

@article{sun2024genesis,
  title={OS-Genesis: Automating GUI Agent Trajectory Construction via Reverse Task Synthesis},
  author={Sun, Qiushi and Cheng, Kanzhi and Ding, Zichen and Jin, Chuanyang and Wang, Yian and Xu, Fangzhi and Wu, Zhenyu and Jia, Chengyou and Chen, Liheng and Liu, Zhoumianze and others},
  journal={arXiv preprint arXiv:2412.19723},
  year={2024}
}