this is an implementation of variational autoencoder using pytorch-lightning. there are some helper classes for a lightning datamodule (with support for tran-validation split) and lightning model.
after training with a 2d latent space, the latent space looks like this:
