Submitted by **Individual-Cause-616** t3_10m4l0b
in **MachineLearning**

I know there is a mathematical way to show that the two approaches of score matching models and diffusion models are the same. I wonder, if there in practice/code are the same either? I already tried to find some PyTorch implementations of score based models but didn’t find anything yet - just for diffusion models.

royalemate357t1_j60xuup wrotethere's an implementation of score-based models from the paper that showed how score based models and diffusion models are the same here: https://github.com/yang-song/score_sde_pytorch

imo their implementation is more or less the same as a diffusion model, except score based models would use a numerical ODE/SDE solver to generate samples instead of using the DDPM based sampling method. it might also train on continuous time, so rather than choosing t ~ randint(0, 1000) it would be t ~ rand_uniform(0, 1.)