ShowUI

Open-source, End-to-end, Lightweight, Vision-Language-Action model for GUI Agent & Computer Use.

ShowUI 是一款开源的、端到端、轻量级的视觉-语言-动作模型，专为 GUI 智能体设计。

   📑 Paper    | 🤗 Hugging Models   |    🤗 Spaces Demo    |    📝 Slides    |    🕹️ OpenBayes贝式计算 Demo
🤗 Datasets   |   💬 X (Twitter)   |    🖥️ Computer Use    |    📖 GUI Paper List    |    🤖 ModelScope

ShowUI: One Vision-Language-Action Model for GUI Visual Agent
Kevin Qinghong Lin, Linjie Li, Difei Gao, Zhengyuan Yang, Shiwei Wu, Zechen Bai, Weixian Lei, Lijuan Wang, Mike Zheng Shou
Show Lab @ National University of Singapore, Microsoft

🔥 Update

• [2025.2.27] ShowUI has been accepted to CVPR 2025.
• [2025.2.13] Support vllm inference.
• [2025.1.20] Support Navigation tasks: Mind2Web, AITW, Miniwob training and evaluator.
• [2025.1.17] Support API Calling via Gradio Client, simply run python3 api.py.
• [2025.1.5] Release the ShowUI-web dataset.
• [2024.12.28] Update GPT-4o annotation recaptioning scripts.
• [2024.12.27] Update training codes and instructions.
• [2024.12.23] Update showui for UI-guided token selection implementation.
• [2024.12.15] ShowUI received Outstanding Paper Award at NeurIPS2024 Open-World Agents workshop.
• [2024.12.9] Support int8 Quantization.
• [2024.12.5] Major Update: ShowUI is integrated into OOTB for local run!
• [2024.12.1] We support iterative refinement to improve grounding accuracy. Try it at HF Spaces demo.
• [2024.11.27] We release the arXiv paper, HF Spaces demo and ShowUI-desktop.
• [2024.11.16] showlab/ShowUI-2B is available at huggingface.

🤖 vllm Inference

See inference_vllm.ipynb for vllm inference.

To leverage multiple GPUs for faster inference, you can adjust the gpu_num parameter

⚡ API Calling

Run python3 api.py by providing a screenshot and a query.

Since we are based on huggingface gradio client, you don’t need a GPU to deploy the model locally 🤗

🖥️ Computer Use

See Computer Use OOTB for using ShowUI to control your PC.

https://github.com/user-attachments/assets/f50b7611-2350-4712-af9e-3d31e30020ee

⭐ Quick Start

See Quick Start for local model usage.

🤗 Local Gradio

See Gradio for installation.

🚀 Training

Our Training codebases supports:

• Grounding and Navigation training: Mind2Web, AITW, Miniwob
• Self-customized model: ShowUI, Qwen2VL
• Efficient Training: DeepSpeed, BF16, QLoRA, SDPA / FlashAttention2, Liger-Kernel
• Multiple datasets mixed training
• Interleaved data streaming
• Image randomly resize (crop, pad)
• Wandb training monitor
• Multi-GPUs, Multi-nodes training

See Train for training set up.

🕹️ UI-Guided Token Selection

Try test.ipynb, which seamless support for Qwen2VL models.

✍️ Annotate your own data

Try recaption.ipynb, where we provide instructions on how to recaption the original annotations using GPT-4o.

❤ Acknowledgement

We extend our gratitude to SeeClick for providing their codes and datasets.

Special thanks to Siyuan for assistance with the Gradio demo and OOTB support.

🎓 BibTeX

If you find our work helpful, please kindly consider citing our paper.

@misc{lin2024showui,
      title={ShowUI: One Vision-Language-Action Model for GUI Visual Agent}, 
      author={Kevin Qinghong Lin and Linjie Li and Difei Gao and Zhengyuan Yang and Shiwei Wu and Zechen Bai and Weixian Lei and Lijuan Wang and Mike Zheng Shou},
      year={2024},
      eprint={2411.17465},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2411.17465}, 
}

If you like our project, please give us a star ⭐ on GitHub for the latest update.