Training compute comparison with diffusion baselines

[viditagr]

Hi, thanks for the great paper! I see that the paper makes a strong case for inference-side efficiency, but I was wondering what was the training cost as compared to diffusion model baselines.