Model metadata not updated between training

There’s currently an issue in the Studio where model metadata (like the confusion matrix and loss) are not updated when you retrain a Neural Network (Keras) block, and have no changed layers (like changing the number of training cycles). We’ll roll out a fix today.

In the meantime the confusion matrix and accuracy shown in Model testing and Deployment are still correct, and you can get the loss and accuracy from the logs:

This has now been resolved!