---
title: graph-cad
canonical_url: "https://www.modelscope.cn/models/JackeySmile/graph-cad"
md_url: "https://www.modelscope.cn/models/JackeySmile/graph-cad.md"
repository: JackeySmile/graph-cad
chinese_name: Graph-CAD
last_updated: 2026-03-27
base_model_relation: finetune
parameters: 523.8M
tensor_type:
  - F32
library_name:
  - safetensors
  - pytorch
frameworks:
  - Pytorch
downloads: 66
stars: 2
---

# graph-cad

> graph-cad - JackeySmile 在 ModelScope 开源的模型。Graph-CAD is a graph-mediated Text-to-CAD model that generates executable Blender bpy code from natural language via a hierarchical and geometry-aware intermediate graph, improving geometric fidelity and constraint…

- **Repository**: JackeySmile/graph-cad
- **Parameters**: 523.8M
- **Downloads**: 66
- **Stars**: 2
- **Last updated**: 2026-03-27

Source: https://www.modelscope.cn/models/JackeySmile/graph-cad

---

# Graph-CAD Model Weights

This repository hosts the released model weights for **Graph-CAD**, a graph-mediated Text-to-CAD framework introduced in:

> **Learning Hierarchical and Geometry-Aware Graph Representations for Text-to-CAD**  
> ICLR 2026

## Description

Graph-CAD tackles long-horizon Text-to-CAD generation by introducing a hierarchical and geometry-aware graph as an intermediate representation. Instead of directly generating executable CAD code from natural language, Graph-CAD first predicts a structured decomposition graph, then converts it into CAD actions and finally into executable Blender `bpy` code.

This ModelScope repository is intended to store the released Graph-CAD model weights and related configuration files.

## Expected Contents

You may place released files such as:

- base model metadata or references
- Graph-CAD fine-tuned checkpoints
- tokenizer/config files when needed
- inference-related configuration files

## Suggested Weight Organization

```text
Graph-CAD-weights/
|- stage1/
|- stage2/
|- stage3/
`- README.md
```

## Usage

After downloading the weights, place them in your local Graph-CAD project under:

```text
checkpoints/stage1
checkpoints/stage2
checkpoints/stage3
```

Then run inference from the Graph-CAD project:

```bash
python infer_api.py
```

## Related Links

- Project repository: `https://github.com/EESJGong/Graph-CAD`
- Base model: `https://modelscope.cn/models/Qwen/Qwen3-8B`

## Citation

If you find Graph-CAD useful in your research, please consider citing:

```bibtex
@inproceedings{gonglearning,
  title={Learning Hierarchical and Geometry-Aware Graph Representations for Text-to-CAD},
  author={Gong, Shengjie and Peng, Wenjie and Chen, Hongyuan and Zhang, Gangyu and Hu, Yunqing and Zhang, Huiyuan and Huang, Shuangping and Chen, Tianshui},
  booktitle={The Fourteenth International Conference on Learning Representations}
}
```