Generative AI Lab 8.3.2

 

Generative AI Lab 8.3.2 focuses on improving support for air-gapped environments, strengthening license validation, and addressing application stability issues.

Additional bug fixes address Models Hub availability, Universal License recognition, air-gapped deployment stability, and file imports.

Improvements

Audit Logging Support for Azure Marketplace Deployments

What’s Improved

Azure Marketplace deployments now include an option to enable audit logging directly during deployment, simplifying the configuration of Elasticsearch and Kibana for audit log management.The deployment process has been updated to include a dedicated checkbox that allows customers to enable audit logging when deploying Generative AI Lab through Azure Marketplace.

Technical Details

  • Deployment Configuration: A new checkbox in the Azure Marketplace deployment interface allows customers to enable audit logging.
  • Elasticsearch and Kibana Support: The deployment process now supports the Elasticsearch and Kibana components required for audit log management.
  • Simplified Setup: Customers can enable audit logging during deployment without requiring additional manual configuration.

Bug Fixes

Models Hub and Online Model Catalog

  • Internal Server Error on the Online Model Page

    The Online Model page could fail to load due to an Internal Server Error when accessing the /api/modelshub/models endpoint. The underlying API issue has been addressed, restoring access to the online model catalog.

  • Models Not Listed in Models Hub

    Models could fail to appear in the Models Hub because the model listing API encountered a KeyError when the compatible_editions field was missing from model metadata. Model listing now handles this condition, allowing available models to be displayed correctly.

Licensing and Air-Gapped Deployments

  • Application License Not Recognized with a Valid Universal License

    Deployments using a valid Universal License could incorrectly display an “Application License Missing” error, preventing normal application operations. The issue also affected Universal License imports in offline mode. Valid licenses are now recognized correctly, allowing users to import licenses, perform application operations, and access licensed pre-annotation and OCR functionality according to their configuration.

  • Pods Failing to Restart in Air-Gapped Environments

    Application pods could fail to restart when an air-gapped instance was stopped and started again. Pod startup and recovery have been corrected, allowing services to return to a stable state after instance restarts.

  • Docker Image Compatibility in Air-Gapped EKS Deployments

    Docker image issues could interfere with successful installation and startup in air-gapped Amazon EKS environments. Deployment images have been updated to support successful installation and application access.

  • Image Mismatches in Air-Gapped Deployments

    Image mismatches affecting air-gapped deployments have been addressed to improve deployment compatibility and reliability.

  • Broken Image References After the Apache Airflow Upgrade

    Image references affected by the Apache Airflow upgrade could interfere with air-gapped installations. The references have been corrected, restoring successful installation and application startup.

Task Import

  • Non-DICOM File Imports Failing Through the Import Server

    The Import Server could process DICOM files while failing to import other supported file types, including PDF documents using Docling-based extraction. Import functionality has been restored for supported non-DICOM files, allowing users to import documents through the Import Server.

Versions

Last updated