What's New
Major features and updates for the Cloudera AI data service.
May 20, 2025
Release notes and fixed issues for version 2.0.50-b65.
New Features / Improvements
- AI Studios (Technical Preview): Cloudera AI Studios is a comprehensive suite of low-code tools designed to simplify the development, customization, and deployment of generative AI solutions within enterprises. This suite empowers organizations to operationalize AI workflows quickly and efficiently by leveraging real-time enterprise data. For more information, see Managing AI Studios.
- Added APIs within the workbench to list Cloudera AI Inference service applications and their associated model endpoints.
- Added Azure UDR support for Cloudera AI Inference service.
- Added Azure NTP support.
- Added API support to retry the creation of Cloudera AI Inference service application upon failure.
- Added a new set of models in the Model Hub, including Llama3.3, DeepSeek-R1-Distill-Llama, Starcorder2, Llama-Nemotron-Nano, NeMo-Retriever-Parse, Llama 3.2 Embedding, and Llama 3.2 Encoder models. To access these models, you must upgrade your Cloudera AI Registries.
- Added support for nim-cli in the AI Registry to import the latest offerings from NVIDIA.
- Enhanced troubleshooting by surfacing underlying issues encountered during AI Registry installation in the Event logs.
- Provided the ability to upgrade the AI Registry directly through the UI, eliminating the reliance on the CLI.
- Implemented automatic redirection to the model import status page whenever a new model import is triggered.
- Users must upgrade their Cloudera AI Inference service applications to serve the latest optimized models from NVIDIA, including Llama3.3, DeepSeek-R1-Distill-Llama, Starcorder2, Llama-Nemotron-Nano, NeMo-Retriever-Parse, Llama 3.2 Embedding, and Llama 3.2 Encoder models.
- Optimization profile details for deployed model endpoints are now surfaced in the UI for improved visibility.
- A user-friendly warning message will now be displayed when replicas of a deployed model scale down.
- Added an option in the UI to retry the creation of Cloudera AI Inference service applications.
- Users will be automatically redirected to the model endpoint page upon triggering the deployment of a new endpoint.
- Enhanced the UI with a variety of user-friendly tooltips for better usability.
- The metrics page for model endpoints will now refresh automatically every 15 seconds for real-time updates.
- GPU count is now auto-selected for NIM profiles when deploying a model endpoint.
- Ensured that dangling pods of deleted endpoints are immediately terminated, preventing them from being left for garbage collection cleanup.
Fixed Issues
- Resolved an issue where duplicate machine user CRNs were preventing the catalog page backup from loading. (DSE-43729)
- Fixed an invalid error issue in the Cloudera AI Registry search filter within the workbench. (DSE-44401)
- Resolved the issues causing failures during the retry of upgrade operations. (DSE-44761)
Apri 28, 2025
Release notes and fixed issues for version 2.0.50-b52.
New Features / Improvements
- Previously, when a job was already running and another job run was triggered by a cron job or an API call, the new run would be skipped and displayed as Failed in the UI. This update introduces a Skipped status, and any skipped job runs will now appear with the Skipped status in the UI.
- The Shared Memory Limit set under Project Settings now applies to both applications and sessions. Previously, it was applied only for sessions.
- Custom Spark settings can now be configured at Cloudera AI Workbench level. When set, the custom Spark configuration provided by the administrator will be merged with the default Spark configuration used in Cloudera AI sessions. These settings will automatically apply to all newly launched Spark sessions within the workbench. The configuration option is available under Site Administration > Runtimes.
- Cloudera AI Workbench now utilizes EBS SSD gp3 volumes for newly created or restored Cloudera AI Workbench instances, replacing the previously used EBS SSD gp2 volumes.
- Added support for Sweden Central on Azure.
- Support for manually modified PVC sizes.
- Added support for EKS 1.31.
- Added support for AKS 1.31.
- The Cloudera AI UI now displays clear error messages for failed Model Imports, enabling quicker troubleshooting.
- Users without the appropriate roles now see actionable error messages in the Model Hub popup.
- Load Balancer Subnet option is added during the AI Registry creation.
- The Cloudera AI UI now supports the force deletion of AI Registry.
- Informative tooltips have been added to the Create Model Endpoints page to improve the user experience.
- Cloudera AI Inference service can now be created without the need for a node group.
- Project template files updated to fully support ML Runtimes. Project template files no longer work with Legacy Engines.
Fixed Issues
- Previously, in the Workbench editor, the search input retained the previous search string when reopened using Ctrl+F or Command+F. This issue has now been resolved. Now, clicking anywhere in the editor text area after a search clears the input, allowing a new search string to be entered. (DSE-40874)
- Increased GRPC operator timeout to ensure reliable handling of more than 150 concurrent session creation attempts. (DSE-36922)
- The Custom Data Connections feature is now properly enabled when the necessary entitlement is present. (DSE-42821)
- Previously, invalid entries in the
runtime_addons
table prevented the registration of models from Cloudera AI Workbenches. This issue has now been resolved. (DSE-44211)
- CVE Fixes – This release includes numerous security fixes for critical and high Common Vulnerability and Exposures (CVE) and includes a CVE on a security vulnerability in the third-party SAML authentication service.
- TLS certificates are now properly deleted during the removal of Cloudera AI Inference service workflow. (DSE-41405)
- Grafana charts have been updated to eliminate references to deprecated metrics. (DSE-42389)
- Cloudera AI Workbench restoration now supports EFS backups larger than 10 TB. (DSE-42986)
- Users can now download the kubeconfig file for the AI Registry as soon as the underlying cluster is provisioned. (DSE-42566)
- Hugging Face token is accurately passed by the UI during Hugging Face model imports. (DSE-42578)
- Instance groups for Cloudera AI Inference service can now be successfully deleted from the UI. (DSE-43182)
- Cloudera AI Inference service no longer overwrites underlying error messages and displays the actual error. (DSE-43315)
- Cloudera AI Inference service UI now accurately reflects the correct state when a user cancels the action of adding a new instance group. (DSE-43183)
- Metrics are now properly displayed in the UI for the model endpoint names that include mixed-case characters. (DSE-43339)
- The root volume size of node groups is now displayed in the Cloudera AI Inference service UI, enabling users to make informed decisions. (DSE-40603)
- Previously, the HTML code generated to embed an image in PBJ Workbench did not work. Now, you can embed images generated in PBJ Workbench-based sessions similarly to how you can embed images from Workbench-based sessions using the share icon next to the generated images. (DSE-42595)
- Previously, the Interrupt button did not work in Cloudera AI Sessions that ran a PBJ Workbench Runtime. This issue has been fixed. (DSE-42344)
-
PBJ Workbench Runtime images now comply with the value set in the MAX_TEXT_LENGTH environmental variable. This limits the maximum number of characters that can be displayed by each command executed. (DSE-42962)
-
Sessions using PBJ-based custom Runtimes with a custom editor could not start previously. This issue has been fixed. (DSE-43297)
-
When PBJ Workbench R Runtime was used, the tables and help text were not properly displayed. This issue has been fixed. (DSE-42077)
-
Previously, when PBJ Workbench R Runtime was used, the editor was not able to show code completion. This issue has been fixed. (DSE-42345)
-
Previously, when PBJ Workbench R Runtime was used, it logged the start of each command execution with a DEBUG severity instead of the INFO severity. This issue has been fixed. (DSE-35299)
-
Previously, when PBJ Workbench R Runtime was used, the content of the tables was not properly displayed. This issue has been fixed. (DSE-42967)
-
Previously, when PBJ Workbench R Runtime was used, the output of the session commands was not properly displayed. This issue has been fixed. (DSE-42966)
Behavioral Changes
- The new Cloudera AI Workbenches will no longer have the Legacy Engine image configured. If you want to configure and use Legacy Engines in your projects, configure them by going to the Site Administration > Runtimes page. (DSE-42593 and DSE-39531)
March 28, 2025
Release notes and fixed issues for version 2.0.47-b365.
Fixed Issues
- The issue of sessions and pods getting stuck in the Stopping state has been resolved. (DSE-42144)
- Pods in an Error or Stuck state within Cloudera AI Workbenches are now being properly garbage-collected. (DSE-43549)
- Reduced the frequency of initialization failures for user workloads that launch immediately after node autoscaling. (DSE-43311)
- Previously, users with MLAdmin roles were initially assigned the MLUser role during the first sync, but their permissions are updated correctly in subsequent syncs or when they log in. This issue is now resolved. (DSE-42775)