2021.10.0

AI Center On-Premises 2021.10.0 LTS Release

We are excited to release AI Center On-Premises 2021.10.0 LTS!

New Features

ML Skills Streaming Logs

You can now check the logs for your ML Skills in near real-time using the Streaming Logs feature. For more information on this, see ML Skills Streaming logs.

Pipelines and ML Skill Infra Settings

You can now customize the resources used for ML Skill and Pipeline while deploying them using the user interface. For more information on this, see:

• Creating pipelines
• Creating ML Skills

Improvements

Installation

It is now easier to install AI Center on-premises. Starting with 2021.10, you now have multiple ways in which you can install AI Center on-premises:

• AI Center standalone
• AI Center in Automation Suite

It is also easier to install ML Packages in an offline environment. For more information, see: ML Packages Offline Installation

Known Issues

• Prediction count on the ML Skills details is not supported in the AI Center existing cluster scenario.
• On rare occasions, if you restart the machine two times consecutively, service deployment can get stuck because of the DATABASECHANGELOGLOCK lock not being released by one service. In this case you will see AI Center pods restarting continuously. If this happens, see the Troubleshooting section.
• Public endpoints for Datasets are not available from activity upload. The public endpoints for Datasets can still be used with HTTP calls directly.
• When using AI Center in the Automation Suite offline environment, some components are not loaded properly. This happens when opening the AI App page from a machine with no internet access.

Document Understanding

• When working in Data Manager, sometimes you may encounter the session not found error message after you logout and log back in. The suggested workaround is to delete the cookies in the browser and refresh the page.
• If you are using 21.10.10 version models and have just upgraded to the newest build, you may experience issues with the prediction feature in Data Manager. The workaround is the redeploy the ML skill.