{"repo":"xlang-ai/OSWorld-G","free":true,"listed":false,"github":"https://github.com/xlang-ai/OSWorld-G","clone":"git clone https://github.com/xlang-ai/OSWorld-G.git","description":"[NeurIPS 2025 Spotlight] Scaling Computer-Use Grounding via UI Decomposition and Synthesis","language":"TypeScript","stars":177,"topics":["agent","benchmark","gui","large-action-model","multimodal","natural-language-processing","rpa","vlm","dataset","models"],"license":"Apache-2.0","category":"workflow-automation","readme_excerpt":"Website • Paper • OSWorld-G Benchmark • Jedi-3B • Jedi-7B • Jedi Dataset (4 million) This is the official repository for \"Scaling Computer-Use Grounding via UI Decomposition and Synthesis\", which includes the benchmark--OSWorld-G and dataset pipeline--Jedi. We also provide links to the models (Jedi-3B, Jedi-7B) and dataset (Jedi) here. 📢 Updates - 2025-05-19: Initial release of this repository. 💾 Environment First, clone this repository and cd into it. Then, install the dependencies listed in requirements.txt . We recommend using the latest version of Conda to manage the environment, but you can also choose to manually install the dependencies. Please ensure that Python version is = 3.9. 🤖 Model To use our model, we recommend using vllm . You need to carefully follow the computer use agent template from Qwen-2.5-VL, and be very careful with the image size to enable the best performance. We show a small example in demo.py You'll get the predicted coordinates of the click position, and the visualization of the click position will be saved as click visualization.png like below: 📊 Benchmark--OSWorld-G We provide our OSWorld-G benchmark with original instructions ( benchmark/OSWorld-G.json ) and refined instructions ( benchmark/OSWorld-G refined.json ) (pure grounding tasks that require minimal additional knowledge). The benchmark data and pipeline code are available in the benchmark folder, along with a series of evaluation scripts in the evaluation folder. For instructions o","default_branch":null,"files":null,"tree":[],"storefront":"/r/xlang-ai","claimed":false,"request_supported":{"post":"https://gitbuyer.com/r/xlang-ai/OSWorld-G/request-supported","requests":0},"note":"indexed from public GitHub; nothing is for sale on this page. Clone it from GitHub. Paid listings live at /search."}