User Guide

This MLflow Plugin allows using the Exasol Bucket File System (BucketFS) as an MLflow Artifact Store.

BucketFS is a powerful feature for exchanging non-relational data with the database nodes in an Exasol cluster.

• MLflow users can store and retrieve artifacts of AI models in the BucketFS of an Exasol database via the MLflow UI or its APIs.

• Applications and User Defined Scripts (UDFs) can retrieve models from Exasol BucketFS very fast. Loading a 600 MB Huggingface model from BucketFS can take less than 2% of the time compared to MLflow’s regular HTTP interface.

Contents

• Installing the Plugin
• Using the BucketFS Artifact Store
• Creating an MLflow Experiment
• URI Format of Artifact URIs
• When and for Which Operations is the MLflow Plugin Required?
• Using a Script Language Container
• Accessing Artifacts from Within a UDF