Determined is an open-source deep learning training platform that makes building models fast and easy.
This is an initial, rough, proof of concept of Determined pre-installed and configured running on a WSL image.
It is not officially supported. However, it does work, and supports NVIDIA CUDA devices.
Install WSL 2:
wsl.exe --install
Download the latest release of determined-wsl.
Import the WSL image:
wsl.exe --import determined-wsl C:\determined-wsl install.tar.gz --version=2
If you prefer a GUI to manage your WSL distributions, I recommend Raft WSL.
From PowerShell:
wsl.exe -d determined-wsl
From Windows Terminal:
determined-wsl will automatically appear in the drop-down box of Windows Terminal.
build.sh - Builds a minimal Ubuntu 22.04 base image with the dependencies needed for Determined AI. Requires Ubuntu 22.04 or later.
config.sh - Is run inside the base image at build time to install Docker and Determined.
config-db.sh - Is run inside the base image at build time as the postgres user to configure PostgreSQL, because sudo and runuser break in chroots and I didn't want to set up systemd-nspawn for this yet.
run.sh - Is run on start of determined-wsl to check PostgreSQL, Docker, Determined services are running, then launches the Determined AI web interface in the default Windows browser. Note: The default credentials for the Determined web UI is admin and no password.
install.ps1 - Checks for WSL updates, imports determined-wsl into WSL 2, and then runs determined-wsl. Note: If WSL is updated then the script will exit and need to be re-run.
- Improve the run.sh script to do more sanity checks and give a nice HUD experence
- Build a nice GUI around this, e.g. Docker Desktop
- Wrap the image in an Win32 installer with Determined icon and WIndows Terminal theme
- Scoop and Winget packages
- Automate fresh image builds using GitHub Actions
- Push the build to the Microsoft Store