A barebones stable diffusion implementation on MNIST

• Implemented as an excercise to gain an intuition as to how stable diffusion works under the hood
• Missing many features such as guided diffusion, cross attention etc.

Commands

• python main.py train --epochs=3
• python main.py inference --n_steps=10