Using EDM for end-to-end training of latent diffusion models

[pandekan]

It would be great if I could please get some guidance on how to adapt this framework for end-to-end training of the parameters of VAE and diffusion models on the VAE's latent space as in this LSGM paper by Vahdat et al. https://github.com/NVlabs/LSGM.