Hopsworks 3.7 The GenAI Release

Hopsworks 3.7 is now generally available. This new release adds additional capabilities to support LLM/GenAI use cases, introduces feature monitoring, integrates support for Delta Lake and provides a new notification service to track changes to specific features and updates its underlying online data store to use the latest version of RonDB.

Feature Monitoring

Deploying a model in production is not the end of the journey. As time passes the world changes, the model will start making inaccurate predictions and new data is available to re-train the model. Hopsworks 3.7 introduces new feature monitoring capabilities to monitor predictions and feature data to detect changes in the underlying data compared to what the model was trained on. Hopsworks will trigger alerts when that happens helping data scientists determine when models need to be re-trained.

Learn more about feature monitoring

Vector Embeddings and Vector Similarity Search Support

Hopsworks 3.7 adds support for approximate nearest neighbor (ANN) search of vector embeddings stored in the feature store. This new functionality enables users to more quickly develop RAG (Retrieval Augmented Generation) pipelines for large language model use cases by integrating in a RAG pipeline both unstructured data (retrieved using the new vector similarity search functionality) as well as structured data. Users can now define a feature group and indicate which feature(s) contains vector embeddings to be indexed for vector similarity search and Hopsworks will index them in an OpenSearch index.

Delta Lake Support

Hopsworks 3.7 adds support for Delta Lake as file format for feature groups. This new capability improves Hopsworks support for Databricks based workflow which can now use a Delta to read and write to the Hopsworks feature store. 

Learn more about Delta support in Hopsworks

Notification Service

With the release of 3.7, Hopsworks introduces a new service which allows a developer to be notified when feature data related to a particular feature group has been changed, for example new data has been written to the feature group or existing data has been updated.

Providing best in class performance for a feature store requires ongoing development and testing work, with the 3.7 release of Hopsworks, the underlying database, RonDB, of the online data store has been upgraded to the latest version taking advantage of significant performance improvements in both throughput and storage, better support for different cloud provider virtual machine types and improved cloud scaling support.

Version Upgrades

This release includes several upgrades of services in Hopsworks. Most notably, Hopsworks 3.7 Python clients now support Python 3.12.