This repo contains minimalistic deep generative models that can be used as starting points when build generative models. Each model is implemented using the MNIST dataset and runs using the PyTorch library. You will also find some of the theory behind the models in these notebooks. Hopefully this is just enough to understand what's going on, no more, no less. If you want the notebooks to run in a reasonable time you will want to use a GPU and have cuda availble. The models covered so far are:
- Variations Autoencoders (VAEs)
- Diffusion Models (DDPMs)
- Generative Adversarial Networks (GANs)
- Autoregressive Models (PixelCNNs)
- Flow Matching