TIGER-Lab/BrowserAgent-RFT

1---2license: apache-2.03datasets:4- TIGER-Lab/BrowserAgent-RFT-Data5language:6- en7base_model:8- Qwen/Qwen2.5-7B-Instruct9metrics:10- success_rate11- efficiency12- safety_score13tags:14- agent15- browser16- web17- rft18---19 20## Model21We release the **RFT (Reward Fine-Tuned)** model used in **BrowserAgent**, initialized from the SFT checkpoint of `Qwen/Qwen2.5-7B-Instruct`.  22This model further optimizes browsing trajectories with task-level reward signals that encourage higher success rate, shorter action paths, and safer interactions.23 24## Paper25[BrowserAgent: Building Web Agents with Human-Inspired Web Browsing Actions](https://arxiv.org/abs/2510.10666)26 27## Project Page28<https://tiger-ai-lab.github.io/BrowserAgent/>29 30## Code31<https://github.com/TIGER-AI-Lab/BrowserAgent>32 33## Sample Usage34```bash35hf download TIGER-Lab/BrowserAgent-RFT --local-dir ./models/browseragent-rft --repo model36```37 38## Citation39```bibtex40@misc{yu2025browseragentbuildingwebagents,41      title={BrowserAgent: Building Web Agents with Human-Inspired Web Browsing Actions}, 42      author={Tao Yu and Zhengbo Zhang and Zhiheng Lyu and Junhao Gong and Hongzhu Yi and Xinming Wang and Yuxuan Zhou and Jiabing Yang and Ping Nie and Yan Huang and Wenhu Chen},43      year={2025},44      eprint={2510.10666},45      archivePrefix={arXiv},46      primaryClass={cs.CL},47      url={https://arxiv.org/abs/2510.10666}, 48}49```