v1.37

Release notes

We are pleased to announce the release of v1.37.0, which includes several new features, improvements, and bug fixes.

✨ New features

• Advanced prediction filtering
  We added new filters for your prediction logs. Filter based on evaluations, actuals or other prediction log attributes in the Predictions tab, or use the new API query parameters to query your prediction logs

• GPU support (Enterprise)
  You can now select GPU nodes for your Deployment. Set up available GPU nodes during installation, and then simply check the customized resources checkbox to select the GPU node that should be used for your model, explainer or transformer.

• Environment variables and arguments
  Using the Python Client or API, you can now pass environment variables and arguments to your models when using KServe as your Deployment backend

🗑️ Dropped support

• featureLabels are no longer a supported metadata.json key. Please use features as mentioned in the metadata.json documentation instead

🛠️ Improvements

• Added bins to the traffic monitoring graphs to align it with other monitoring graphs
• Provide information in a tooltip when the standard explainer is disabled
• When using the API, your default custom resources are properly configured
• Maintain the monitoring settings when switching between tabs

🐛 Bug fixes

• Fixed an event's pod log message when a node is unavailable
• Fixed an issue with the bin size on the drift monitoring graph
• Fixed an issue where you'd be able to click on explainer and transformer options in a Deployment not unauthorized to do so