wjbmattingly/NuMarkdown-8B-Thinking-Demo

🤖 NuMarkdown-8B Reasoning Demo

A Gradio-based web application that demonstrates the reasoning capabilities of the NuMarkdown-8B model from NumInd. This app allows users to upload images and see both the model's detailed thinking process and final analysis.

🌟 Features

• —Visual Analysis: Upload any image for AI analysis
• —Reasoning Transparency: See the model's step-by-step thinking process
• —Clean Interface: Side-by-side layout with tall frames for better visibility
• —Zero GPU Integration: Optimized for HuggingFace Spaces with GPU acceleration
• —Real-time Processing: Automatic analysis when images are uploaded

🚀 Quick Start

Option 1: Deploy to HuggingFace Spaces (Recommended)

1. 1.Create a new Space on HuggingFace Spaces
2. 2.Choose:
3. 3.SDK: Gradio
4. 4.Hardware: Zero GPU (for best performance)
5. 5.Upload these files to your Space:
6. 6.app.py
7. 7.requirements.txt
8. 8.README.md
9. 9.Your Space will automatically build and deploy!

Option 2: Run Locally

# Clone this repository
git clone <your-repo-url>
cd NuMarkdown-8B-Thinking-Demo

# Install dependencies
pip install -r requirements.txt

# Run the application
python app.py

The app will be available at http://localhost:7860

🔧 Technical Details

Model Information

• —Model: numind/NuMarkdown-8B-reasoning
• —Type: Vision-Language Model with reasoning capabilities
• —Framework: Qwen2.5-VL architecture
• —Features: Structured thinking with <think> and <answer> tags

Dependencies

• —Gradio 4.44.0: Web interface framework
• —PyTorch: Deep learning framework
• —Transformers: HuggingFace model library
• —Flash Attention 2: Optimized attention mechanism
• —Spaces: HuggingFace Zero GPU integration

📱 How to Use

1. 1.Upload an Image: Click on the image upload area on the left side
2. 2.Wait for Processing: The model will automatically analyze your image
3. 3.View Results:
4. 4.Reasoning Panel: See the model's detailed thinking process
5. 5.Answer Panel: Get the final conclusion or analysis

🎯 Use Cases

• —Document Analysis: Analyze text, tables, charts, and diagrams
• —Educational Content: Understand complex visual information
• —Research: Extract insights from academic papers and figures
• —General Vision: Describe and analyze any visual content

🛠️ Customization

Modify Generation Parameters

In app.py, you can adjust:

model_output = model.generate(
    **model_input, 
    temperature=0.7,      # Creativity level (0.1-1.0)
    max_new_tokens=5000   # Maximum response length
)

UI Customization

The interface uses custom CSS for tall frames. Modify the css parameter in gr.Blocks() to adjust the layout.

📊 Performance Notes

• —Zero GPU Spaces: Provides the best performance for this model
• —Memory Requirements: ~16GB VRAM recommended for optimal performance
• —Processing Time: Typically 10-30 seconds depending on image complexity

🤝 Contributing

Feel free to submit issues and enhancement requests!

📄 License

This project is open source. Please check the license of the underlying model (numind/NuMarkdown-8B-reasoning) for commercial use restrictions.

🙏 Acknowledgments

• —NumInd for the amazing NuMarkdown-8B-reasoning model
• —HuggingFace for the Transformers library and Spaces platform
• —Gradio for the easy-to-use web interface framework

Built with ❤️ for the AI community