README.md

---
base_model:
- Qwen/Qwen2.5-Coder-7B-Instruct
datasets:
- luzimu/webgen-agent_train_step-grpo
- luzimu/webgen-agent_train_sft
license: mit
pipeline_tag: image-text-to-text
library_name: transformers
---

# WebGen-Agent

WebGen-Agent is an advanced website generation agent designed to autonomously create websites from natural language instructions. It was introduced in the paper [WebGen-Agent: Enhancing Interactive Website Generation with Multi-Level Feedback and Step-Level Reinforcement Learning](https://arxiv.org/pdf/2509.22644v1).

Code: https://github.com/mnluzimu/WebGen-Agent

## Project Overview

WebGen-Agent combines state-of-the-art language models with specialized training techniques to create a powerful website generation tool. The agent can understand natural language instructions specifying appearance and functional requirements, iteratively generate website codebases, and refine them using visual and functional feedback.

## Resources

Links to the data and model parameters are as follows:

| **Data** | **HF Link** |
|----------|------|
| **webgen-agent_train_sft** | 🤗 [luzimu/webgen-agent_train_sft](https://huggingface.co/datasets/luzimu/webgen-agent_train_sft) |
| **webgen-agent_train_step-grpo** | 🤗 [luzimu/webgen-agent_train_step-grpo](https://huggingface.co/datasets/luzimu/webgen-agent_train_step-grpo) |

| **Model** | **HF Link** |
|----------|------|
| **WebGenAgent-LM-7B-SFT** | 🤗 [luzimu/WebGenAgent-LM-7B-SFT](https://huggingface.co/luzimu/WebGenAgent-LM-7B-SFT) |
| **WebGenAgent-LM-7B-Step-GRPO** | 🤗 [luzimu/WebGenAgent-LM-7B-Step-GRPO](https://huggingface.co/luzimu/WebGenAgent-LM-7B-Step-GRPO) |
| **WebGenAgent-LM-8B-SFT** | 🤗 [luzimu/WebGenAgent-LM-8B-SFT](https://huggingface.co/luzimu/WebGenAgent-LM-8B-SFT) |
| **WebGenAgent-LM-8B-Step-GRPO** | 🤗 [luzimu/WebGenAgent-LM-8B-Step-GRPO](https://huggingface.co/luzimu/WebGenAgent-LM-8B-Step-GRPO) |

## How WebGen-Agent Works

WebGen-Agent follows an iterative, multi-step paradigm for website generation:

1. **Code Generation**: The agent generates code to create or edit website files based on natural language instructions
2. **Code Execution**: Dependencies are installed and the website service is started
3. **Feedback Gathering**: 
   - A screenshot of the website is captured
   - A Visual Language Model (VLM) provides appearance feedback and scores
   - A GUI-agent tests the website functionality and provides functional feedback
4. **Refinement**: Based on the feedback, the agent continues to improve the website until it meets requirements

![WebGen-Agent Workflow](fig/webgen-agent.png)

## Step-GRPO with Screenshot and GUI-agent Feedback

The Step-GRPO with Screenshot and GUI-agent Feedback approach uses the screenshot and GUI-agent scores inherently produced in the WebGen-Agent workflow as step-level rewards:
- **Screenshot Score**: Quantifies the visual appeal and aesthetics of the website
- **GUI-agent Score**: Measures how well the website meets functional requirements

These dual rewards provide dense, reliable process supervision that significantly improves the model's ability to generate high-quality websites.

![Step-GRPO with Screenshot and GUI-agent Feedback](fig/step-grpo.png)

## Sample Usage

Before running inference, you should rename `.env.template` to `.env` and set the base urls and api keys for the agent-engine LLM and feedback VLM. They can be obtained from any openai-compatible providers such as [openrouter](https://openrouter.ai/), [modelscope](https://www.modelscope.cn/my/overview), [bailian](https://bailian.console.aliyun.com/#/home), and [llmprovider](https://llmprovider.ai/).

You can also deploy open-source VLMs and LLMs by running `src/scripts/deploy_qwenvl_32b.sh` and `src/scripts/deploy.sh`. Scripts for single inference and batch inference can be found at `src/scripts/infer_single.sh` and `src/scripts/infer_batch.sh`.

```bash
python src/infer_single.py \
    --model deepseek-chat \
    --vlm_model Qwen/Qwen2.5-VL-32B-Instruct \
    --instruction "Please implement a wheel of fortune website." \
    --workspace-dir workspaces_root/test \
    --log-dir service_logs/test \
    --max-iter 20 \
    --overwrite \
    --error-limit 5
```

## Citation

If you find our project useful, please cite:

```
@misc{lu2025webgenagentenhancinginteractivewebsite,
      title={WebGen-Agent: Enhancing Interactive Website Generation with Multi-Level Feedback and Step-Level Reinforcement Learning}, 
      author={Zimu Lu and Houxing Ren and Yunqiao Yang and Ke Wang and Zhuofan Zong and Junting Pan and Mingjie Zhan and Hongsheng Li},
      year={2025},
      eprint={2509.22644},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2509.22644}, 
}

@misc{lu2025webgenbenchevaluatingllmsgenerating,
      title={WebGen-Bench: Evaluating LLMs on Generating Interactive and Functional Websites from Scratch}, 
      author={Zimu Lu and Yunqiao Yang and Houxing Ren and Haotian Hou and Han Xiao and Ke Wang and Weikang Shi and Aojun Zhou and Mingjie Zhan and Hongsheng Li},
      year={2025},
      eprint={2505.03733},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2505.03733}, 
}
```