mbrede/notebook-gpu repository overview

⁠HAW Kiel JupyterHub Notebook Images

Custom notebook images for the JupyterHub teaching infrastructure at HAW Kiel (Hochschule für Angewandte Wissenschaften Kiel). Built for use with JupyterHub on a Kubernetes (RKE2) cluster and designed for courses in Data Science, Cloud Computing, and Generative AI.

⁠Images

⁠Kernels

All images expose three Jupyter kernels:

• Python (uv) — Python 3.12 in a uv⁠-managed virtual environment
• R — R via r2u⁠ binary packages for Ubuntu 24.04
• Marimo — Marimo⁠ reactive notebook kernel

⁠Python Packages

All images include the following Python packages (installed via uv into a virtual environment at ~/.venv):

jupyterlab, jupyterhub, notebook, ipykernel, ipympl, ipywidgets, altair, beautifulsoup4, bokeh, bottleneck, cloudpickle, dask, dill, h5py, jupyterlab-git, matplotlib, numba, numexpr, numpy, openpyxl, pandas, patsy, plotly, scikit-image, scikit-learn, scipy, seaborn, sentence-transformers, sqlalchemy, statsmodels, sympy, transformers, xlrd, jupyter-resource-usage, marimo, uv

The notebook-cpu image adds CPU-only PyTorch. The notebook-gpu image adds CUDA-enabled PyTorch and jupyterlab-nvdashboard for GPU monitoring.

⁠R Packages

R packages are installed via r2u⁠ as pre-compiled Ubuntu binaries (significantly faster than install.packages()):

r-base, caret, crayon, devtools, e1071, forecast, hexbin, htmltools, htmlwidgets, IRkernel, nycflights13, randomForest, RCurl, rmarkdown, RODBC, RSQLite, shiny, tidymodels, tidyverse

⁠Attribution

The startup scripts (start.sh, run-hooks.sh, fix-permissions), server configuration (jupyter_server_config.py), and container conventions (jovyan user, NB_UID/NB_GID remapping, before-notebook.d hooks) are adapted from the Jupyter Docker Stacks⁠ project.

The following files are copied verbatim from jupyter/docker-stacks:

• fix-permissions (from docker-stacks-foundation)
• run-hooks.sh (from docker-stacks-foundation)
• start.sh (from docker-stacks-foundation)
• start-singleuser.py (from base-notebook)
• docker_healthcheck.py (from base-notebook)

The following files are adapted:

• start-notebook.py — prepends the uv venv to PATH before delegating to start.sh
• jupyter_server_config.py — CONDA_DIR reference replaced with VENV for SSL certificate path resolution

⁠Key Differences from Jupyter Docker Stacks

• No conda/mamba — Python package management is handled entirely by uv⁠
• uv venv as kernel — the active Python kernel is a uv-managed venv at ~/.venv, not the conda base environment
• R via r2u — binary R packages from Ubuntu apt instead of conda
• Built from ubuntu:24.04 — no conda base layer

⁠Source

github.com/MBrede/haw-kiel-jupyter⁠