Manage artifacts in AI Model Experiments

Secure artifact management is essential for preserving models, datasets, and plots throughout the ML lifecycle. STACKIT AI Model Experiments leverages STACKIT Object Storage for this purpose. Since the storage is S3-compatible, you can interact with your buckets using familiar tools like boto3, the AWS CLI, or the native MLflow™ client.

This guide will walk you through setting up and managing your artifact storage (S3) for use with MLflow™ on the STACKIT platform. For more details, please refer to Object Storage.

1. Locate Your storage bucket

When you create an AI Model Experiments instance, STACKIT automatically generates a default bucket for you inside your project. This bucket is pre-configured to sync with the MLflow™ UI. While you have the flexibility to connect any other STACKIT Object Storage bucket to your experiments, please note that artifacts stored in custom buckets will not be accessible or visible through the MLflow™ UI.

1. Log in to the STACKIT Portal.
2. In the left-hand navigation menu, select Object Storage.
3. Select Buckets.
4. Identify the pre-generated bucket. Its name is listed in the instance details (see Manage Instances).

2. Configure access credentials

To allow the Python MLflow™ client to interact with the artifact storage, you must provide valid access credentials. Upon the creation of an AI Model Experiment instance, STACKIT automatically generates a dedicated credential group named aiexp-{bucket-id}-client. You can generate new credentials directly within this group; however, you are also free to use the default credential group or a custom one, provided it has sufficient permissions to access the bucket.

1. Within the Object Storage section, navigate to Credentials & Groups.
2. Select a credential group.
3. On the left, click on Credentials.
4. Click on Create credentials.
   Once the credentials (Access Key ID and Secret Access Key) appear in the popup, copy and save them immediately. For security reasons, they cannot be viewed again once the window is closed. If you lose them, you will have to delete the credential and create a new one.