Topics

#ai #amd64 #arm64 #container #cuda #data-science #devcontainer #docker #gpu #julia #jupyterlab #machine-learning #ml #mojo #multi-arch #nvidia #podman #python #r #vscode

Language breakdown

No language data available.

▸ Files click to expand

README

(CUDA-based) Data Science dev containers

(GPU accelerated) Multi-arch (linux/amd64, linux/arm64/v8) Data Science dev containers:

(CUDA) Julia base, pubtools
MAX/Mojo base, scipy
CUDA MAX base, scipy
(CUDA) Python base, scipy
(CUDA) R base, tidyverse, verse, geospatial, qgisprocess

Dev containers considered stable for

Julia versions ≥ 1.7.3
Mojo versions ≥ 24.3.0

* MAX versions ≥ 24.5.0

Python versions ≥ 3.10.5
R versions ≥ 4.2.0

Parent images

(CUDA-based) Julia docker stack
(CUDA-based) MAX/Mojo docker stack
(CUDA-based) Python docker stack
(CUDA-based) R docker stack

Extended to match the (CUDA-based) Jupyter Lab docker stacks, except that

GPU accelerated dev containers are based on the

NVIDIA CUDA runtime flavoured image. * The JupyterLab docker stacks are based on the NVIDIA CUDA devel flavoured image.

Dev containers' Oh My Zsh uses the devcontainers theme + default font.

* The JupyterLab docker stacks' Oh My Zsh uses Powerlevel10k theme + MesloLGS NF font.

Features

JupyterLab: A web-based interactive development environment for Jupyter

notebooks, code, and data.

Git: A distributed version-control system for tracking changes in source

code.

Git LFS: A Git extension for versioning large files.
GRASS GIS: A free and open source Geographic Information System (GIS).

:information_source: (CUDA) R qgisprocess image

Orfeo Toolbox: An open-source project for state-of-the-art remote

sensing. :information_source: (CUDA) R qgisprocess image (amd64 only)

Julia[^1]: A high-level, high-performance dynamic language for technical

computing.

MAX[^1]: A high-performance generative AI framework.
Mojo[^1]: A programming language for AI developers.
Pandoc: A universal markup converter.
Python: An interpreted, object-oriented, high-level programming language

with dynamic semantics.

QGIS: A free, open source, cross platform (lin/win/mac) geographical

information system (GIS). :information_source: (CUDA) R qgisprocess image

Quarto: A scientific and technical publishing system built on Pandoc.

:information_source: Julia pubtools, MAX/Mojo/Python scipy, R verse+ images

R[^1]: A language and environment for statistical computing and graphics.
arf: An alternative R frontend — a modern R console written in Rust.
SAGA GIS: A Geographic Information System (GIS) software with immense

capabilities for geodata processing and analysis. :information_source: (CUDA) R qgisprocess image

TinyTeX: A lightweight, cross-platform, portable, and easy-to-maintain

LaTeX distribution based on TeX Live. :information_source: Julia pubtools, MAX/Mojo/Python scipy, R verse+ images

Zsh: A shell designed for interactive use, although it is also a powerful

scripting language.

:point_right: See the Version Matrices for detailed information:

Julia: Version Matrix,

CUDA Version Matrix

MAX/Mojo: Version Matrix,

CUDA Version Matrix

Python: Version Matrix,

CUDA Version Matrix

R: Version Matrix,

CUDA Version Matrix

Pre-installed extensions

:information_source: Pinned to version 11.7.0 due to unsolicited AI content

:information_source: Julia pubtools, MAX/Mojo/Python scipy, R verse+ images

[^1]: Depending on which dev container configuration is selected.

Prerequisites

Dev containers require either Docker or Podman[^2] to be installed. CUDA-based versions require the following in addition:

NVIDIA GPU
NVIDIA Linux driver
NVIDIA Container Toolkit

[^2]: See issue #1 about limitations in Podman.

:information_source: The host running the GPU accelerated dev containers only requires the NVIDIA driver, the CUDA toolkit does not have to be installed.

Use driver version 580 (Long Term Support Branch) with NVIDIA Data Center GPUs or select NGC-Ready NVIDIA RTX boards to ensure forward compatibility until June 2028.

Install

Codespaces require no installation, but do not currently offer machines with NVIDIA GPUs.

Docker

To install Docker, follow the instructions for your platform:

Podman

To install Podman, follow the instructions for your platform:

Podman Installation | Podman

CUDA

To install the NVIDIA Container Toolkit, follow the instructions for your platform:

Installation Guide — NVIDIA Cloud Native Technologies documentation

Usage

The default dev container is meant to work on this repository.

Every other configuration is a custom Data Science dev container that behaves in a unique way:

Default mount[^3]:

* source: empty directory * target: /home/vscode * type: volume

Codespace only mount:

* source: root of this repository * target: /workspaces * type: misc

Default path: /home/vscode
Default user[^4]: vscode

* uid: 1000 (auto-assigned) * gid: 1000 (auto-assigned)

Lifecycle scripts:

* onCreateCommand: home directory setup * postStartCommands * docker: Silently remove all unused images and all build cache (Codespace only) * julia: Copy user-specific startup files * r: Copy QGIS stuff from skeleton directory; Create R user library * postAttachCommand: Codespace only: Check for dev container updates

[^3]: See issue #2 about changing the mount type.

[^4]: See issue #3 about running as root.

To disable the postStartCommand or postAttachCommand, comment out line 8 in ~/.local/bin/dockerSystemPrune.sh or ~/.local/bin/checkForUpdates.sh.

Codespace

Click the <> Code button, then click the Codespaces tab.

A message is displayed at the bottom of the dialog telling you who will pay for the codespace.

Create your codespace after configuring advanced options:

* Configure advanced options To configure advanced options for your codespace, such as a different machine type or a particular devcontainer.json file: * At the top right of the Codespaces tab, select ... and click New with options.... * On the options page for your codespace, choose your preferred options from the dropdown menus. * Click Create codespace.

– Creating a codespace for a repository - GitHub Docs

To open your codespace in JupyterLab:

Execute

jupyter-lab \ --ServerApp.allow_origin='*' \ --ServerApp.cookie_opti \ --ServerApp.tornado_settings="{'headers':{'Content-Security-Policy':\"frame-ancestors 'self' https://*.github.dev\", 'Access-Control-Allow-Headers': 'accept, content-type, authorization, x-xsrftoken, x-github-token'}}" \ --notebook-dir=/home/vscode \ --no-browser

Ctrl+click on one of the URLs shown in the Terminal.

:information_source: Opening your codespace in JupyterLab according to the GitHub Docs sets the default path to /workspaces/<repository-name> that you can not escape.

Local/'Remote SSH'

Use the Dev Containers: Reopen in Container command from the Command Palette (F1, ⇧⌘P (Windows, Linux Ctrl+Shift+P))

To start JupyterLab:

Execute

jupyter-lab

Ctrl+click on one of the URLs shown in the Terminal.

Persistence

Data in the following locations is persisted:

The user's home directory (/home/vscode)[^5]
The dev container's workspace (/workspaces)

[^5]: Alternatively for the root user (/root). Use with Docker/Podman in rootless mode.

This is accomplished either via a volume or bind mount (or loop device on Codespaces) and is preconfigured.

| Codespaces: A 'Full Rebuild Container' resets the home directory!
:information_source: This is never necessary unless you want exactly that. | |:----------------------------------------------------------------------------------------------------------------------------------------------------|