An artifact is any data that is produced and/or consumed by functions, jobs, or pipelines.

Artifacts metadata is stored in the project’s database. The main types of artifacts are:

  • Files — files, directories, images, figures, and plotlines

  • Datasets — any data, such as tables and DataFrames

  • Models — all trained models

  • Feature Store Objects — Feature Sets and Feature Vectors

In this section

Viewing artifacts#

Artifacts can be viewed and managed in the UI. In the project page, select the artifact type models, files, or feature-store (for datasets and feature store objects).

Example dataset artifact screen:


You can search the artifacts based on time and labels. For each artifact, you can view its location, the artifact type, labels, the producer of the artifact, the artifact owner, last update date, and type-specific information.

Artifacts can also be viewed from the Jobs > Artifacts tab. For each artifact, you can view its content as well as download the artifact.

Artifact path#

Jobs use the default or job specific artifact_path parameter to determine where the artifacts are stored. The default artifact_path can be specified at the cluster level, client level, project level, or job level (at that precedence order), or can be specified as a parameter in the specific log operation.

You can set the default artifact_path for your environment using the set_environment() function.

You can override the default artifact_path configuration by setting the artifact_path parameter of the set_environment() function. You can use variables in the artifacts path, such as {{project}} for the name of the running project or {{run.uid}} for the current job/pipeline run UID. (The default artifacts path uses {{project}}.) The following example configures the artifacts path to an artifacts directory in the current active directory (./artifacts)

set_environment(project=project_name, artifact_path='./artifacts')

For Iguazio MLOps Platform users

In the platform, the default artifacts path is a /artifacts directory in the predefined “projects” data container: /v3io/projects/<project name>/artifacts (for example, /v3io/projects/myproject/artifacts for a “myproject” project).

Saving artifacts in run-specific paths#

When you specify {{run.uid}}, the artifacts for each job are stored in a dedicated directory for each executed job. Under the artifact path, you should see the source-data file in a new directory whose name is derived from the unique run ID. Otherwise, the same artifacts directory is used in all runs, and the artifacts for newer runs override those from the previous runs.

As previously explained, set_environment returns a tuple with the project name and artifacts path. You can optionally save your environment’s artifacts path to a variable, as demonstrated in the previous steps. You can then use the artifacts-path variable to extract paths to task-specific artifact subdirectories. For example, the following code extracts the path to the artifacts directory of a training task, and saves the path to a training_artifacts variable:

from os import path
training_artifacts = path.join(artifact_path, 'training')


The artifacts path uses data store URLs, which are not necessarily local file paths (for example, s3://bucket/path). Be careful not to use such paths with general file utilities.

Artifact URIs, versioning, and metadata#

Artifacts have unique URIs in the form store://<type>/<project>/<key/path>[:tag]. The URI is automatically generated by log_artifact and can be used as input to jobs, functions, pipelines, etc.

Artifacts are versioned. Each unique version has a unique IDs (uid) and can have a tag label.
When the tag is not specified, it uses the latest version.

Artifact metadata and objects can be accessed through the SDK or downloaded from the UI (as YAML files). They host common and object specific metadata such as:

  • Common metadata: name, project, updated, version info

  • How they were produced (user, job, pipeline, etc.)

  • Lineage data (sources used to produce that artifact)

  • Information about formats, schema, sample data

  • Links to other artifacts (e.g. a model can point to a chart)

  • Type-specific attributes

Artifacts can be obtained via the SDK through type specific APIs or using generic artifact APIs such as:

Example artifact URLs:


See also#

Back to top