Building an variational autoencoder with transformers

Hi, I would like to build a variational autoencoder using a pretrained transformer model from huggingface.
I have seen a few papers doing this but could not find a code example.
Is it even possible to use transformers for VAEs? I don’t really see how one could generate random samples from the learned latent representation with a transformer decoder.
If it is possible, could someone link me to a code example?