I\'m working on variational autoencoder and my loss function returns Nan value. Yes i have tf.log and i added 1e-10 to avoid log(0). I am using TF 1.12. Here is my loss func