nanoDDPM

中文

Train a tiny diffusion model even with CPU. Implemented with Pytorch. Tiny U-net style model and mnist dataset using only label 0 and 1

---

[TOC]

Prepare

1. python3.8
2. pip install -r requirements.txt

Eval

Pretrained model trained with 20000 iterations. Result save in eval_out More iterations should be better. Strongly recommend you try it.

bash eval.sh

Train

bash train.sh

Feel free to tune the training arguments as you want. Have fun!

Welcome to pull request.