PyTorch-VAE

Why we included this project

Researchers and ML engineers looking for a working reference on variational autoencoders will find this collection genuinely useful. It bundles many well-known VAE variants, among them beta-VAE, VQ-VAE, IWAE, WAE, conditional and categorical VAEs, and keeps the architectures structurally similar so the differences between models are easy to inspect. Since every model trains on the same CelebA data with consistent layer choices, the published reconstructions and samples give a fair baseline when you are weighing an architecture for your own generative task. The code is organized around simple YAML configs and PyTorch Lightning, so you can swap in your own dataset or adjust latent dimensions without rewriting model logic. Treat it as a reproducible study guide for paper results rather than a production pipeline; it is a strong starting point, not a drop-in deployment system.