Errors in training VAE with Tensorflow and switching to PyTorch

Question

I\'m trying to implement a variational autoencoder that aims to perform unsupervised classification (and possibly generation) of astronomical images. Each image is of size 63x63