Model Serving via Unity Catalog

[zsucic1]

Hi everyone! Has anyone successfully deployed a model saved on Unity Catalog to Model Serving? I get:Event Log: Served model creation failed for served model 'model', config version 15. Error message: Container creation failed. Please see build logs for more information.
Service Log: There are currently no replicas in a running state.
Build Log: Build never started - check the event log to see if the model failed validation or contact Databricks.I can deploy the same model via Workspace MLflow. The only interesting thing I see on Unity Catalogue Model UI is the invisible signature:

Failed to load model version signature. Error: Access to the storage container is forbidden by Azure.

[Kaniz]

Hi @zsucic1, It seems you’re encountering some issues while deploying a model saved in Unity Catalog to Model Serving. Let’s troubleshoot this together.

Access to Storage Container:

• The error message indicates that there’s an issue with accessing the storage container due to permissions. Ensure that the service account or identity you’re using has the necessary permissions to read from the storage container where the model artifacts are stored.
• Verify that the credentials used for accessing the storage account are correct and have the required permissions.

Build Logs and Validation:

• Check the build logs as suggested in the error message. These logs might provide more detailed information about why the build process failed.
• Also, review the validation process for the model. If the model failed validation, it could prevent successful deployment. Ensure that the model adheres to the required format and dependencies.

Invisible Signature:

• The invisible signature issue might be related to the model version signature. This could be due to incorrect permissions or misconfiguration.
• Double-check the configuration settings for the Unity Catalog and ensure that the model version signature is accessible.

Unity Catalog vs. Workspace MLflow:

• You mentioned that you can deploy the same model via Workspace MLflow. Compare the configurations and settings between the two deployment methods.
• Ensure that the Unity Catalog setup aligns with the requirements for model serving.

Dependencies and Environment:

• Verify that all necessary dependencies (such as Python libraries) are correctly specified for the model. Unity Catalog relies on these dependencies during deployment.
• If you’re using custom Python libraries, consider uploading them to Unity Catalog volumes or Databri....

 If you encounter any specific issues, feel free to share additional details, and we’ll continue troubleshooting! 🛠️🔍

[sachinw]

We are facing the same issue.