• NVIDIA-themed dark mode with green accent colors (#76B900)
• Automatic notebook opening from cloned repositories
• GPU monitoring dashboards (when GPUs available)
• Interactive guided tours
• Remote kernel gateway support
• Disabled update/news notifications
• Multi-CUDA support: Pre-built images for CUDA 11.8 and 12.1
• OpenShift compatible: Works with arbitrary UIDs (uses aggressive 777 permissions for guaranteed compatibility)

Environment variables:

• GITHUB_REPO: Optional Git repo to clone into /home/jovyan/work/repo.
• AUTO_NOTEBOOK: Optional path (relative to repo root or absolute) of a notebook to auto-open.
• AUTO_NOTEBOOK_GLOB: Optional glob (e.g., *.ipynb) to pick the first matching notebook.
• JUPYTER_TOKEN: If empty, token auth is disabled. If set, will be used as login token.
• KERNEL_GATEWAY: If set to 1, start jupyter kernelgateway (headless kernel mode) instead of JupyterLab.
• KERNEL_GATEWAY_PORT: Port for kernel gateway (default 9999).
• NOTEBOOK_URL: Optional URL to download notebook from. Supports hybrid pattern: Kubernetes orchestrator uses initContainer (preferred), standalone Docker downloads directly (fallback).

Branding:

• Dark theme is default with NVIDIA accent (#76B900).
• Favicon shows NVIDIA logo.
• Top-left logo replaced with "NVIDIA" text.
• Green accent colors throughout UI.

GPU dashboards:

• jupyterlab-nvdashboard is installed. Access via the "GPU Dashboards" menu in JupyterLab.
• Note: GPU dashboards only appear when NVIDIA GPUs are detected. Use --gpus all flag with Docker.
• If no GPUs are available, the extension loads but the menu won't appear.

Interactive tour:

• JupyterLab Tour extension is installed and auto-starts on first launch.
• Two guided tours available:
  • Welcome to NVIDIA JupyterLab: Overview of features and navigation
  • GPU Features Tour: How to access GPU dashboards and monitoring
• Access tours anytime from Help menu > Tours

Remote kernel usage from a lightweight client:

• Start the container with -e KERNEL_GATEWAY=1 -p 9999:9999.
• From a local JupyterLab client, install jupyter_server_gateway and configure a Gateway Server at http://<host>:9999.

This project supports multiple CUDA versions via separate base images:

• CUDA 11.8: For older GPUs and legacy compatibility (V100, T4, RTX 20xx/30xx)
• CUDA 12.1: Latest stable for modern GPUs (A100, H100, L4, RTX 40xx)

Pull pre-built images from GHCR:

# Latest CUDA 12.1
docker pull ghcr.io/[your-org]/pyrrhus-jupyter:latest

# Specific CUDA versions
docker pull ghcr.io/[your-org]/pyrrhus-jupyter:cuda-11.8
docker pull ghcr.io/[your-org]/pyrrhus-jupyter:cuda-12.1

Build all CUDA variants:

cd lp-pyrrhus-jupyter-launch
./build-cuda.sh

Build a specific CUDA version:

./build-cuda.sh 12.1

Or manually with Docker:

docker build -f Dockerfile.cuda \
  --build-arg CUDA_VERSION=12.1 \
  --build-arg UBUNTU_VERSION=22.04 \
  -t pyrrhus-jupyter:cuda-12.1 .

Using pre-built CUDA images:

# Launch with repo clone and auto-open first notebook
docker run --gpus all -p 8888:8888 \
  -e GITHUB_REPO=https://github.com/brevdev/notebooks.git \
  ghcr.io/[your-org]/pyrrhus-jupyter:cuda-12.1

# Launch with explicit notebook
docker run --gpus all -p 8888:8888 \
  -e GITHUB_REPO=https://github.com/brevdev/notebooks.git \
  -e AUTO_NOTEBOOK=nemo-reranker.ipynb \
  ghcr.io/[your-org]/pyrrhus-jupyter:cuda-12.1

# Launch with notebook from URL (standalone Docker pattern)
docker run --gpus all -p 8888:8888 \
  -e NOTEBOOK_URL=https://raw.githubusercontent.com/brevdev/launchables/main/biomistral.ipynb \
  -e AUTO_NOTEBOOK=biomistral.ipynb \
  ghcr.io/[your-org]/pyrrhus-jupyter:cuda-12.1

# Launch with pre-mounted notebook (volume mount pattern)
docker run --gpus all -p 8888:8888 \
  -v $(pwd)/my-notebook.ipynb:/home/jovyan/work/my-notebook.ipynb \
  -e AUTO_NOTEBOOK=/home/jovyan/work/my-notebook.ipynb \
  ghcr.io/[your-org]/pyrrhus-jupyter:cuda-12.1

# Headless kernel mode (remote kernel)
docker run --gpus all -p 9999:9999 \
  -e KERNEL_GATEWAY=1 \
  ghcr.io/[your-org]/pyrrhus-jupyter:cuda-12.1

Using locally built images:

# Standard build (non-CUDA)
docker build -t nv-jlab -f Dockerfile .

# Launch with repo clone
docker run --gpus all -p 8888:8888 \
  -e GITHUB_REPO=https://github.com/brevdev/notebooks.git \
  nv-jlab

GitHub Actions automatically builds and pushes images to GHCR on:

• Push to main branch
• Git tags (e.g., v1.0.0)
• Manual workflow dispatch

Images are tagged with:

• cuda-{version} - Specific CUDA version
• cuda-{version}-latest - Latest build of that CUDA version
• cuda-{version}-{git-sha} - Specific commit
• latest - Latest CUDA version (12.1)