This educational repository provides a clean implementation of Denoising Diffusion Probabilistic Models (DDPM) using PyTorch. It includes training scripts for MNIST, CIFAR-10, and CelebA datasets, with a simple UNet architecture and time embeddings. Users can run the self-contained superminddpm.py script directly or train on larger datasets with train_cifar10.py. The implementation generates images by iteratively denoising random samples, following the forward/backward diffusion process.