Cloudera Docs

What's New

Major features and updates for the Cloudera AI data service.

August 22, 2025

Release notes and fixed issues for version 2.0.52-b34.

New Features / Improvements

ML Runtimes
  • Released new Spark Runtime Addons versions - 2.4.8, 3.2.3, 3.3.0, and 3.5.1.
  • Spark 3.5.1 in Cloudera AI is now certified to work with the Data Lake 7.3.1 version.
  • New HadoopCLI Runtime Addon versions 7.2.18 and 7.3.1 are now available. Versions 7.2.16 and 7.2.17 have been removed.

Fixed Issues

Cloudera AI Registry
  • Resolved an issue that blocked AI Registry creation in regions other than us-west-2. (DSE-46535)
  • Resolved an issue that prevented registries in renew:finished status from being used by Cloudera AI Inference service. Upgrading Cloudera AI Inference service is required to use these registries. (DSE-46815)
  • Previously, the UI would flicker when trying to access the Registered Models UI due to an issue. This issue is now resolved. (DSE-46714)
  • Resolved a critical bug that caused cdh-client to restart in a workbench, which prevented sessions from starting. (DSE-46989)
Cloudera AI Workbench
  • Previously, users were unable to register models from within Cloudera AI Workbenches. This issue is now resolved. Upgrading workbenches is required to register models from within Cloudera AI Workbenches . (DSE-46656)

July 31, 2025

Release notes and fixed issues for version 2.0.52-b27.

New Features / Improvements

Cloudera AI Platform
  • Added support for file storage replication in AWS EFS (Elastic File System), enhancing data redundancy and availability. For information, see Configuring File Storage Replication on AWS.
  • Improvements have been made to enable retriable in-place upgrades, leading to more robust upgrade processes.
  • For information, see Upgrading Cloudera AI Workbenches.
  • Added support for Azure regions: Poland Central and Italy North.
  • Added support for Istio.
  • Added support for Customer-Managed Key (CMK) encryption in Cloudera AI Azure workbenches. For information, see Enabling Customer Managed Keys on Microsoft Azure.
  • You can now change Persistent Volume Claim (PVC) sizes through both the UI on the Workbench Details page and using the CDP CLI. For information, see Modifying workbench persistent volume size.
  • Added a liveness probe to the mlx crud application pod and implemented graceful shutdown, improving the stability and resilience of the application.
  • The system will now automatically select the Cloudera AI Registry if only one instance exists within a given tenant.
  • Improvements have been made to the UI, allowing you to create Cloudera AI Registries with private clusters and enable User-Defined Routing (UDR) more easily.
  • Added user-friendly information to the UI to assist users when utilizing the air-gapped model hub import functionality.
Cloudera AI Registry
  • Added CSI driver support for AI Registries, removing previous resource constraints. You can now download any number of models in parallel without encountering resource limitations with Azure.
  • User-friendly and informative error messages are now displayed when users are unable to import a model to the AI Registry.
  • A new caching mechanism has been introduced in Model Hub, significantly reducing the time it takes for pages to load.
Cloudera AI Inference service
  • Added support for the Nemotron Super 49B model.
  • Added support for Riva ASR NIM (NVIDIA Inference Microservice), enabling advanced automatic speech recognition. This feature is compatible with the Whisper mode, requiring a 16-bit, mono, 16000 Hz, uncompressed WAV file as input.
  • Added support for several new vLLM load formats, including sharded_state, gguf, bitsandbytes, mistral, runai_streamer, and fastsafetensors. This enhances the list of supported vLLM quantization options.
  • Nemotron's thinking mode is now user-configurable, allowing you to explicitly activate this advanced reasoning capability by including "content": "detailed thinking on" within the system role of your prompt payload, giving you precise control over resource usage.
  • Implemented necessary validators for GPU instance types during the deployment of NVIDIA models to prevent misconfigurations.
  • Significantly improved the performance of Cloudera AI Inference service by caching tokens to improve UI responsiveness and decrease network load.
  • The replica for endpoint logs and events is now automatically selected for any given model endpoint.
  • Added a Refresh button to various sub-sections of the model endpoint details page for easier data updates.
  • A force fetch button is now available on the Model Hub UI for users to override cached values and ensure the latest data is displayed.
  • Replaced generic Failed to Fetch messages with more user-friendly error messages when a user attempts to import a Hugging Face model not present in our Model Hub.
  • An alert box is now displayed in the UI to notify users when an Ingress Ready endpoint has a replica count of 0.
ML Runtimes
  • Resource requests for several core Cloudera AI services have been increased. This change is designed to boost performance and stability, ensuring a smoother experience without requiring any action on your part.

Fixed Issues

Cloudera AI Platform
  • Resolved an issue that prevented Cloudera AI Registries from being visible in the control plane after their certificates were renewed. (DSE-44836)
  • Previously, the configuration map of Cloudera AI Inference service was not updating correctly during the upgrade process. This issue is now resolved. (DSE-45417)
Cloudera AI Registry
  • Previously, constraints in the UI prevented upgrading an AI Registry already in a Ready status. This issue is now resolved. (DSE-45663)
Cloudera AI Workbench
  • Addressed an issue due to which registries whose certificates were renewed were not visible from within the workbench. (DSE-44837)
Cloudera AI Inference service
  • Resolved an issue that prevented the configuration map of AI Inference from updating correctly during the upgrade process. (DSE-45417

  • Resolved an issue encountered when importing the A10G profile of the Llama 3.2 rerank 1B model. (DSE-45375)

  • Previously, copying the base URL of model endpoints from the UI did not work properly, as the wrong link was getting copied. This issue is now resolved.(DSE-45107 and DSE-45534)
  • Resolved an issue that prevented rendering of Test Model and code samples for external Hugging Face models, such as Gemma 3. (DSE-45419)
  • Addressed inconsistencies observed in the Summary and Details metrics displayed for Model Endpoints. (DSE-45185)
  • Fixed an issue where the Cloudera AI Registry upgrade pop-up was not correctly being dismissed. (DSE-46224).
  • Previously, when the Test Model under Model Endpoint was executed, the UI blocked navigation to other tabs, such as Metrics, Logs, and so on. This issue is now resolved. (DSE-46247)
  • Previously, due to a known KServe issue (kserve/kserve#4471), all newly created model endpoints would initially deploy with a single replica, regardless of the specified configuration. This issue is now resolved. (DSE-45876)
ML Runtimes
  • Previously, PBJ Workbench-based workloads kept running when the underlying kernel had stopped or restarted. This is now fixed, and PBJ Workbench-based workloads will terminate if the underlying kernel terminates. (DSE-42964)
  • Previously, messages printed from PBJ Workbench-based models did not appear in the model logs. This issue is now resolved. (DSE-42960)
  • Previously, in the PBJ Workbench editor where text with special styling (for example, colored fonts) lost its formatting in the console. This issue is now resolved. (DSE-42958)
  • Previously, PBJ Workbench Runtimes did not stop when "exit" was executed in a Python kernel or when "quit()" was executed in an R kernel. This issue is now resolved, and now these commands terminate the workload as expected. (DSE-36835)
  • Fixed the known issue related to Spark executors in R. Now the environment variable R_LIBS_USER has the same configuration in Spark executors as in other workloads. It is no longer needed to manually configure R_LIBS_USER for Spark executors in R. (DSE-32839)
  • Previously, some types of output were not fully shown in the Workbench UI when the workload was running a PBJ Workbench Runtime. This issue is now resolved. (DSE-43865)

June 06, 2025

Release notes and fixed issues for version 2.0.50-b68 and MLX-CRUD-APP version 1.50.0-b139.

New Features / Improvements

Cloudera AI Platform
  • Added support for Azure AMD Easv5 instances. (DSE-38566)