Packages

Artifact-ML comprises three packages:

artifact-core

The framework foundation.

It provides a declarative interface for the computation of validation artifacts in ML experiments.

Its objective is to enable reusable validation workflows by providing the tools to trigger artifacts by name---with zero adapter code.

In line with our design philosophy, achieving this sets the stage for Artifact-ML’s broader objective: the elimination of imperative glue code in ML experiments at large.

For more details consult the package's docs.

artifact-experiment

Experiment orchestration extension.

It provides the tools to build reusable validation workflows with integrated tracking using popular backend services (e.g. Mlflow).

For more details consult the package's docs.

artifact-torch

Pytorch integration.

It offers the tools to build reusable, end-to-end deep learning workflows declaratively.

It handles all aspects of the training loop aside from model architecture and data pipelines, abstracting away engineering complexity to let researchers focus on architectural innovation.

For more details consult the package's docs.

Getting Started

For installation instructions refer to the following page.