Skip to main content

Prerequisites

AIchor has prerequisites for both engine setup and running experiments.

Engine prerequisites​

Creating or importing an AWS, GCP or Azure compute engine requires a cloud provider account with specific permissions (e.g. IAM roles on AWS or Azure RBAC roles on Azure) so AIchor can provision and maintain the necessary infrastructure.

The exact requirements vary depending on the cloud provider and whether the engine is being created or imported. See Engine Technical Requirements for details.

Project prerequisites​

  • An engine needs to be provisioned and in Ready state

  • A VCS provider needs to be set up

For a step-by-step walkthrough, see Creating a Project

Prerequisites for running experiments​

  • A project needs to be created and in Ready state

A Dockerfile. AIchor builds and runs experiments inside a container. The Dockerfile defines the image in the container that will be executed.

A manifest.yaml file at the repository root. This file tells AIchor what resources to allocate, which machine learning framework to use, and what command to run. See the Manifest File Reference for the full specification.

More information on experiments is available at How Experiments Work.

Setting up a local repository​

The AIchor CLI is able to prepare a local repository to run on AIchor. From an empty directory, aichor local-repo init scaffolds a Dockerfile, a manifest.yaml, a pyproject.toml, and a uv-compliant directory tree. See CLI First Steps for the full flow.

For ready-to-run examples covering each supported framework, see the aichor-demo repository.