Tensorflow probability module in production

[Tridib dutta]

Hello all,

I have a model based on TFP (a variational autoencoder). I want to use a JAVA program to read in the trained model saved as .h5 file. However, while loading the saved model, I am having issues with some of the layers (KLDivergenceAddLoss) to be precise. I am wondering if there is a work around or should we avoid TFP to build the model in the first place? 

Any suggestion would be greatly appreciated. 

Thank you.