I have implemented a variational autoencoder with the Keras implementation as an example (https://keras.io/examples/generative/vae/). When plotting the training loss I notic
