How well does this model compare to stable-cascade?

[JohnnyRacer]

This looks like an awesome and very promising project to train cheaper diffusion models. However, it seems like stable-cascade is still vastly more popular and widely used. Are there any apples-to-apples performance/speed/quality comparisons on the PixArt models compared to stable-cascade? It seems like both of these projects are trying to solve the same problem of training scalability, so it would be helpful to have some reproducible metrics between these two models. All of the examples shown in both stable-cascade and PixArt shown seem to be quite cherry-picked so its hard to compare directly.