| datasets: | |
| - hflqf88888/SWIRL_GUI_data | |
| base_model: | |
| - Qwen/Qwen2.5-VL-3B-Instruct | |
| The instantiation of SWIRL's dual-agent architecture in mobile GUI control. The Navigator translates instructions, history, and screenshots into structured low-level instructions (LLI), while the Interactor executes them as atomic actions (click, scroll, text input) with precise grounding. This hierarchical design enhances robustness, generalization, and interpretability. | |
| For more details, please refer to our [project repository](https://github.com/Lqf-HFNJU/SWIRL). |