Spaces:
Running
Running
title: GLM-4.5V CAD Generator | |
emoji: π§ | |
colorFrom: blue | |
colorTo: green | |
sdk: gradio | |
sdk_version: 5.43.1 | |
app_file: app.py | |
pinned: false | |
license: mit | |
hardware: zero-gpu | |
python_version: 3.11 | |
# π§ GLM-4.5V CAD Generator | |
Generate CADQuery Python code from 3D CAD model images using GLM-4.5V vision-language models! | |
## Features | |
- **Multiple GLM-4.5V variants**: AWQ, FP8, and full precision models | |
- **Advanced prompting**: Simple, detailed, and chain-of-thought styles | |
- **Real-time generation**: GPU-accelerated inference on Zero GPU | |
- **Code extraction**: Automatic cleaning and formatting of generated CADQuery code | |
- **Evaluation ready**: Compatible with existing VSR and IoU metrics | |
## Usage | |
1. Upload a clear 3D CAD model image | |
2. Select your preferred GLM-4.5V model variant | |
3. Choose a prompting style | |
4. Click "Generate CADQuery Code" | |
5. Copy the generated Python code and run it locally | |
## Models | |
- **GLM-4.5V-AWQ**: Fastest startup, good quality | |
- **GLM-4.5V-FP8**: Balanced speed and quality | |
- **GLM-4.5V**: Best quality, slower startup | |
## Tips | |
- Use clear, well-lit CAD images for best results | |
- Simple geometric shapes work better than complex assemblies | |
- Try different prompt styles if first attempt isn't satisfactory | |
- Generated code may need minor adjustments for complex geometries | |
## Technical Details | |
This app uses Hugging Face Zero GPU (H200 GPUs, 70GB VRAM) for serverless GPU allocation and the latest GLM-4.5V vision-language models for image-to-code generation. Built with Gradio 5.42.0 for optimal 2025 compatibility and enhanced authentication support. The generated CADQuery code follows standard patterns with proper imports and result variables. |