Kubeflow PyTorch Notebook Container

GPU-accelerated PyTorch development environment optimized for Kubeflow with JupyterLab and SSH access.

Quick Start

In Kubeflow GUI

Notebooks -> + New Container / + New Notebook

ghcr.io/KNU-PLML-Lab/kubeflow-nvcr-pytorch:latest

Docker Compose

# Your SSH public key (optional)
export SSH_PUBLIC_KEY="$(cat ~/.ssh/id_rsa.pub)"
docker-compose up -d

Docker Directly

# with SSH
docker run -d \
  -p 8888:8888 \
  -p 22:22 \
  --gpus all \
  -e SSH_PUBLIC_KEY="$(cat ~/.ssh/id_rsa.pub)" \
  -v $(pwd)/notebooks:/home/jovyan/work \
  ghcr.io/KNU-PLML-Lab/kubeflow-nvcr-pytorch:latest

# Jupyter only
docker run -d \
  -p 8888:8888 \
  --gpus all \
  -v $(pwd)/notebooks:/home/jovyan/work \
  ghcr.io/KNU-PLML-Lab/kubeflow-nvcr-pytorch:latest

Access

• Jupyter Notebook: (http 8888)
• SSH: (port 22)

Environment Variables

Variable	Description	Default
SSH_PUBLIC_KEY	SSH public key for authentication	None
NB_PREFIX	Base URL prefix for Jupyter	/
JUPYTER_ENABLE_LAB	Enable JupyterLab interface	yes

Building from Source

# Clone the repository
git clone https://github.com/KNU-PLML-Lab/kubeflow-nvcr-pytorch.git
cd kubeflow-nvcr-pytorch

# Build the image
./build.sh