Improve model card: Add pipeline tag, library, license, and quick start

README.md

@@ -1,12 +1,58 @@
-### Model Overview
+---
+license: cc-by-nc-4.0
+library_name: transformers
+pipeline_tag: image-text-to-text
+---
+
+# VeriOS: Query-Driven Proactive Human-Agent-GUI Interaction for Trustworthy OS Agents
 
 This model is a **Query-Driven Trustworthy OS Agent** implemented as described in the paper:
+[VeriOS: Query-Driven Proactive Human-Agent-GUI Interaction for Trustworthy OS Agents](https://arxiv.org/abs/2509.07553).
+It is initialized with weights from the **Qwen2.5-VL-32B** model.
 
-**VeriOS: Query-Driven Proactive Human-Agent-GUI Interaction for Trustworthy OS Agents**
+## Quick Start
 
-It is initialized with weights from the **Qwen2.5-VL-32B** model.
+### 1. Environment Setup
+1. Clone the repository:
+   ```bash
+   git clone https://github.com/Wuzheng02/VeriOS
+   ```
+2. Navigate into the project directory:
+   ```bash
+   cd VeriOS
+   ```
+3. Download the VeriOS-Bench dataset:
+  
+   [https://huggingface.co/datasets/wuuuuuz/VeriOS-Bench](https://huggingface.co/datasets/wuuuuuz/VeriOS-Bench)
+
+4. Download the pre-trained models:
+
+   VeriOS-Agent-7B: [https://huggingface.co/wuuuuuz/VeriOS-Agent-7B](https://huggingface.co/wuuuuuz/VeriOS-Agent-7B)
+
+   VeriOS-Agent-32B: [https://huggingface.co/wuuuuuz/VeriOS-Agent-32B](https://huggingface.co/wuuuuuz/VeriOS-Agent-32B)
+
+### 2. Evaluation
+1. Evaluate VeriOS-Agent performance:
+   ```bash
+   python test_interaction_loop.py --model_path /path/to/VeriOS-Agent --json_path /path/to/test.json
+   ```
+2. Evaluate dual-agent system performance:
+   ```bash
+   python dual_agent.py --model_path1 /path/to/scenarioagent --model_path2 /path/to/actionagent --json_path /path/to/test.json
+   ```
+3. Evaluate other baselines:
+   ```bash
+   python test_loop_{name}.py --model_path /path/to/agent --json_path /path/to/test.json
+   ```
+
+### 3. Training
+This work is based on full fine-tuning of LLMs using [LLaMA-Factory](https://github.com/hiyouga/LLaMA-Factory). We gratefully acknowledge the support from the LLaMA-Factory project.
+
+To reproduce the training process of VeriOS-Agent from scratch:
+1. Replace the `.yaml` files in the LLaMA-Factory repository with those provided in this repository.
+2. Follow the official training tutorials provided in the [LLaMA-Factory repository](https://github.com/hiyouga/LLaMA-Factory).
 
-### Citation
+## Citation
 
 ```bibtex
 @article{wu2025verios,
@@ -15,3 +61,4 @@ It is initialized with weights from the **Qwen2.5-VL-32B** model.
   journal={arXiv preprint arXiv:2509.07553},
   year={2025}
 }
+```