Selecting an Image#

Using one of the Jupyter Docker Stacks requires two choices:

  1. Which Docker image you wish to use

  2. How you wish to start Docker containers from that image

This section provides details about the first.

Core Stacks#

The Jupyter team maintains a set of Docker image definitions in the jupyter/docker-stacks GitHub repository. The following sections describe these images, including their contents, relationships, and versioning strategy.

CUDA enabled variant#

We provide CUDA accelerated version of pytorch-notebook image. Prepend a CUDA version prefix (like cuda12-) to the image tag to allow PyTorch operations to use compatible NVIDIA GPUs for accelerated computation. We only build images for 2 last major versions of CUDA.

For example, you can use an image


Source on GitHub | Dockerfile commit history | image tags

jupyter/docker-stacks-foundation is a small image supporting a majority of options common across all core stacks. It is the basis for all other stacks on which Jupyter-related applications can be built (e.g., kernel-based containers, nbclient applications, etc.). As such, it does not contain application-level software like JupyterLab, Jupyter Notebook, or JupyterHub.

It contains:

  • Package managers

    • conda: “cross-platform, language-agnostic binary package manager”.

    • mamba: “reimplementation of the conda package manager in C++”. We use this package manager by default when installing packages.

  • Unprivileged user jovyan (uid=1000, configurable, see options in the common features section of this documentation) in group users (gid=100) with ownership over the /home/jovyan and /opt/conda paths

  • tini and a script as the container entry point - useful for running alternative commands in the container as applications are added (e.g. ipython, jupyter kernelgateway, jupyter lab)

  • A script, which can source/run files in a given directory

  • Options for a passwordless sudo

  • Common system libraries like bzip2, ca-certificates, locales

  • wget to download external files

  • No preinstalled scientific computing packages


Source on GitHub | Dockerfile commit history | image tags

jupyter/base-notebook adds base Jupyter Applications like JupyterLab, Jupyter Notebook, JupyterHub, and NBClassic and serves as the basis for all other stacks besides jupyter/docker-stacks-foundation.

It contains:

  • Everything in jupyter/docker-stacks-foundation

  • Minimally functional Server (e.g., no LaTeX support for saving notebooks as PDFs)

  • notebook, jupyterhub, and jupyterlab packages

  • A script as the default command

  • A script useful for launching containers in JupyterHub

  • Options for a self-signed HTTPS certificate


jupyter/base-notebook also contains and files to maintain backward compatibility. External config that explicitly refers to those files should instead update to refer to and The shim .sh files will be removed at some future date.


Source on GitHub | Dockerfile commit history | image tags

jupyter/minimal-notebook adds command-line tools useful when working in Jupyter applications.

It contains:

  • Everything in jupyter/base-notebook

  • Common useful utilities like curl, git, nano (actually nano-tiny), tzdata, unzip, and vi (actually vim-tiny),

  • TeX Live for notebook document conversion


Source on GitHub | Dockerfile commit history | image tags

jupyter/r-notebook includes popular packages from the R ecosystem listed below:


Source on GitHub | Dockerfile commit history | image tags

jupyter/julia-notebook includes popular packages from the Julia ecosystem listed below:

  • Everything in jupyter/minimal-notebook and its ancestor images

  • The Julia compiler and base environment

  • IJulia to support Julia code in Jupyter notebook

  • Pluto.jl reactive Julia notebook interface, made accessible with jupyter-pluto-proxy

  • HDF5 package


Source on GitHub | Dockerfile commit history | image tags

jupyter/scipy-notebook includes popular packages from the scientific Python ecosystem.


Source on GitHub | Dockerfile commit history | image tags

jupyter/tensorflow-notebook includes popular Python deep learning libraries.

  • Everything in jupyter/scipy-notebook and its ancestor images

  • tensorflow machine learning library


Source on GitHub | Dockerfile commit history | image tags

jupyter/pytorch-notebook includes popular Python deep learning libraries.

  • Everything in jupyter/scipy-notebook and its ancestor images

  • pytorch machine learning library


Source on GitHub | Dockerfile commit history | image tags

jupyter/datascience-notebook includes libraries for data analysis from the Python, R, and Julia communities.

  • Everything in the jupyter/scipy-notebook, jupyter/r-notebook, and jupyter/julia-notebook images and their ancestor images

  • rpy2 package


Source on GitHub | Dockerfile commit history | image tags

jupyter/pyspark-notebook includes Python support for Apache Spark.


Source on GitHub | Dockerfile commit history | image tags

jupyter/all-spark-notebook includes Python and R support for Apache Spark.

  • Everything in jupyter/pyspark-notebook and its ancestor images

  • IRKernel to support R code in Jupyter notebooks

  • rcurl, sparklyr, ggplot2 packages

Image Relationships#

The following diagram depicts the build dependency tree of the core images. (i.e., the FROM statements in their Dockerfiles). Any given image inherits the complete content of all ancestor images pointing to it.

Image inheritancediagram


Every Monday and whenever a pull request is merged, images are rebuilt and pushed to the public container registry.

Versioning via image tags#

Whenever a docker image is pushed to the container registry, it is tagged with:

  • the latest tag

  • a 12-character git commit SHA like 1ffe43816ba9

  • a date formatted like 2023-01-30

  • OS version like ubuntu-22.04

  • a set of software version tags like python-3.10.8 and lab-3.5.3


  • Tags before 2022-07-05 were sometimes incorrect. Please, do not rely on them.

  • Single-platform images have either aarch64- or x86_64- tag prefixes, for example,

For stability and reproducibility, you should either reference a date formatted tag from a date before the current date (in UTC) or a git commit SHA older than the latest git commit SHA in the default branch of the jupyter/docker-stacks GitHub repository.

Community Stacks#

The core stacks are but a tiny sample of what’s possible when combining Jupyter with other technologies. We encourage members of the Jupyter community to create their own stacks based on the core images and link them below. See the contributing guide for information about how to create your own Jupyter Docker Stack.






More than 200 Jupyter Notebooks with example C# code



nbgrader and RISE on top of the datascience-notebook image



Based on IHaskell. Includes popular packages and example notebooks



IJava kernel on top of the minimal-notebook image



sagemath kernel on top of the minimal-notebook image



Major geospatial Python & R libraries on top of the datascience-notebook image



Kotlin kernel for Jupyter/IPython on top of the base-notebook image



Transformers and NLP libraries such as Tensorflow, Keras, Jax and PyTorch



Scraper tools (selenium, chromedriver, beatifulsoup4, requests) on minimal-notebook image



Scala kernel for Jupyter using Almond on top of the base-notebook image

Other GPU-accelerated notebooks#




Power of your NVIDIA GPU and GPU calculations using Tensorflow and Pytorch in collaborative notebooks. This is done by generating a Dockerfile that consists of the nvidia/cuda base image, the well-maintained docker-stacks that is integrated as a submodule, and GPU-able libraries like Tensorflow, Keras and PyTorch on top of it.


PRP (Pacific Research Platform) maintained registry for jupyter stack based on NVIDIA CUDA-enabled image. Added the PRP image with Pytorch and some other Python packages and GUI Desktop notebook based on jupyterhub/jupyter-remote-desktop-proxy.


GPU accelerated, multi-arch (linux/amd64, linux/arm64/v8) docker images for R, Python and Julia. Derived from nvidia/cuda devel-flavored images, including TensortRT and TensorRT plugin libraries. With code-server next to JupyterLab. Just Python – no Conda/Mamba.