Keras tensorflow custom loss using intermediate tensorflow probability layer with Eager execution

Question

When creating a custom loss for a model using intermediate tensorflow probability DistributionLambda layer, it was easy without Eager execution, as the layer could