spark-playbooks

A collection of playbooks and recipes for deploying AI models and workloads on NVIDIA DGX Spark systems.

Overview

This repository contains ready-to-use playbooks for running various AI models and frameworks optimized for NVIDIA DGX Spark GPUs. Each playbook includes Docker configurations, build scripts, and comprehensive documentation.

Available Playbooks

🚀 Nemotron v2 VL on vLLM

Run the Nemotron Nano VL 12B V2 vision-language model using vLLM with FP4 quantization.

• Model: nvidia/Nemotron-Nano-VL-12B-V2-FP4-QAD
• Framework: vLLM (optimized for inference)
• Quantization: FP4 for reduced memory usage
• API: OpenAI-compatible REST API
• Use Cases: Vision-language tasks, multimodal AI applications

📖 View Documentation

Repository Structure

spark-playbooks/
├── README.md                           # This file
├── run-nemotron-v2-VL/                 # Nemotron VL playbook
│   ├── README.md                       # Playbook-specific documentation
│   ├── build.sh                        # Build script
│   ├── run.sh                          # Run script
│   ├── Dockerfile                      # Docker configuration
│   └── patch1.patch                    # Custom patches
└── [other-playbooks]/                  # Additional playbooks
    └── README.md                       # Playbook-specific documentation

Quick Start

1. Choose a playbook from the list above
2. Navigate to the playbook directory:

   cd [playbook-name]

3. Follow the playbook's README.md for detailed instructions

General Prerequisites

All playbooks in this repository require:

• NVIDIA DGX Spark system with GPU support
• Docker installed with NVIDIA Container Toolkit
• NVIDIA GPU drivers (recommended: latest stable version)
• Sufficient disk space (varies by playbook)

Verify Your Setup

Check that your system is ready:

# Check NVIDIA drivers
nvidia-smi

# Check Docker NVIDIA runtime
docker run --rm --gpus all nvidia/cuda:12.0.0-base-ubuntu22.04 nvidia-smi

# Check available disk space
df -h

Contributing

We welcome contributions! To add a new playbook:

1. Create a new directory with a descriptive name
2. Include all necessary files (Dockerfile, scripts, patches)
3. Add a comprehensive README.md in the playbook directory
4. Update this main README.md to list your playbook

Playbook README Template

Each playbook should include:

• Overview: What the playbook does
• Prerequisites: Specific requirements
• Build Instructions: Step-by-step build process
• Run Instructions: How to launch and configure
• Usage Examples: API calls, code samples
• Configuration: Tuning parameters
• Troubleshooting: Common issues and solutions

Additional Resources

• NVIDIA DGX Spark Documentation
• NVIDIA Container Toolkit
• Docker Documentation

Support

For playbook-specific issues, refer to the individual playbook's README.md documentation.

For general DGX Spark system issues, contact NVIDIA support.

License

See LICENSE file for details.