ML Runtimes Environment Variables

This topic describes how ML Runtimes environmental variables work. It also lists the different scopes at which they can be set and the order of precedence that will be followed in case of conflicts.

ML Runtimes environment variables behave the same way for Legacy Engines and ML Runtimes.

Environmental variables allow you to customize ML Runtimes environments for projects. For example, if you need to configure a particular timezone for a project, or increase the length of the session/job timeout windows, you can use environmental variables to do so.

CML allows you to define environmental variables for the following scopes:
Global

A site administrator for your CML deployment can set environmental variables on a global level. These values will apply to every project on the deployment.

To set global environmental variables, go to Admin > Runtime/Engine.

Project

Project administrators can set project-specific environmental variables to customize the ML Runtimes launched for a project. Variables set here will override the global values set in the site administration panel.

To set environmental variables for a project, go to the project's Overview page and click Settings > Advanced.

Job

Environments for individual jobs within a project can be customized while creating the job. Variables set per-job will override the project-level and global settings.

To set environmental variables for a job, go to the job's Overview page and click Settings > Set Environmental Variables.

Experiments

ML Runtimes created for execution of experiments are completely isolated from the project. However, these ML Runtimes inherit values from environmental variables set at the project-level and/or global level. Variables set at the project-level will override the global values set in the site administration panel.

Models

Model environments are completely isolated from the project. Environmental variables for these ML Runtimes can be configured during the build stage of the model deployment process. Models will also inherit any environment variables set at the project and global level. However, variables set per-model build will override other settings.