dgx-spark-demo

DGX Spark Demo

This demo showcases NVIDIA's Nemotron Nano 9B v2 model running locally via NIM (NVIDIA Inference Microservice) on DGX Spark, with OpenWebUI as the frontend interface.

Overview

• Model: NVIDIA Nemotron Nano 9B v2
• Runtime: NVIDIA NIM
• Frontend: OpenWebUI
• Hardware: DGX Spark

Prerequisites

• NVIDIA DGX Spark or compatible GPU hardware
• Docker installed with NVIDIA GPU support
• NGC API Key (Get one here)
• Python 3.x (for OpenWebUI)

Setup Instructions

1. Running Local NIM

First, set up and run the NVIDIA Inference Microservice with the Nemotron model:

# Set your NGC API key
export NGC_API_KEY="<your-ngc-api-key>"

# Set up cache directory
export LOCAL_NIM_CACHE=~/.cache/nim
mkdir -p "$LOCAL_NIM_CACHE"

# Run the NIM container
docker run -it --rm \
    --gpus all \
    --shm-size=16GB \
    -e NGC_API_KEY \
    -v "$LOCAL_NIM_CACHE:/opt/nim/.cache" \
    -u $(id -u) \
    -p 8000:8000 \
    nvcr.io/nim/nvidia/nvidia-nemotron-nano-9b-v2-dgx-spark:latest

The NIM service will be available at http://localhost:8000.

2. Installing OpenWebUI

Follow the official OpenWebUI installation guide:

OpenWebUI Installation Guide

3. Running OpenWebUI

Configure and start OpenWebUI to connect to your local NIM instance:

# Configure OpenWebUI to use the local NIM endpoint
export OPENAI_API_BASE_URL=http://0.0.0.0:8000/v1
export OPENAI_API_KEY=""
export ENABLE_MODEL_SELECTOR=false
export WEBUI_AUTH=False
export DEFAULT_MODEL="nvidia/nemotron-nano-9b-v2"

# Start OpenWebUI
open-webui serve --host 0.0.0.0 --port 8080

Access OpenWebUI at http://localhost:8080 in your browser.

Testing the Setup

To verify that the NIM service is running correctly, use the following curl command:

curl -X 'POST' \
  'http://localhost:8000/v1/chat/completions' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "nvidia/nemotron-nano-9b-v2",
    "messages": [{"role":"user", "content":"Which number is larger, 9.11 or 9.8?"}],
    "max_tokens": 128,
    "stream": true
  }'

You should receive a streaming response from the model.

Architecture

┌─────────────┐     HTTP      ┌──────────────┐     API      ┌─────────────┐
│  User       │ ────────────> │  OpenWebUI   │ ──────────>  │  NIM        │
│  Browser    │               │  (Port 8080) │              │  (Port 8000)│
└─────────────┘               └──────────────┘              └─────────────┘
                                                                    │
                                                                    v
                                                             ┌─────────────┐
                                                             │  Nemotron   │
                                                             │  Nano 9B v2 │
                                                             └─────────────┘

Troubleshooting

NIM Container Issues

• Ensure you have sufficient GPU memory (the model requires significant VRAM)
• Verify your NGC API key is valid
• Check that Docker has access to GPU resources with docker run --gpus all nvidia/cuda:11.0-base nvidia-smi

OpenWebUI Connection Issues

• Verify the NIM service is running: curl http://localhost:8000/v1/models
• Ensure the OPENAI_API_BASE_URL points to the correct endpoint
• Check firewall settings if accessing from a different machine

Additional Resources

• NIM Container for DGX Spark (NGC Catalog)
• NVIDIA NIM Documentation
• OpenWebUI Documentation
• Nemotron Model Information

License

Refer to NVIDIA's licensing terms for NIM and the Nemotron model.