Get the Container Images#
Dioptra runs as a set of Docker containers. Before you can deploy Dioptra, you need container images for the core services (REST API, MLflow tracking, NGINX) and at least one worker image (PyTorch or TensorFlow).
There are two ways to obtain these images:
Which Method Should I Use?#
Download pre-built images if:
You want to get started quickly
The pre-built worker environments meet your needs
Build images locally if:
You want to customize the container images
You are in a restricted environment that cannot access the GitHub Container Registry
Understanding Container Registry Prefixes#
Container images have a registry prefix that identifies where they came from. This prefix affects how you configure your Dioptra deployment.
Method |
Image Name Example |
Registry Prefix |
|---|---|---|
Downloaded from GHCR |
|
|
Built locally |
|
(empty) |
When you prepare your deployment, you will set the container_registry template variable to match your images:
Downloaded images: Leave
container_registryat its default value ofghcr.io/usnistgovLocally built images: Set
container_registryto an empty string
Note
If you need to mix downloaded images with locally built images (e.g., using downloaded images for most services but a custom-built image for a specific worker), you will need to use Docker Compose override files to specify the correct image. See Use Docker Compose Override Files for details.