Rectified flow and SD3

[kkj15dk]

I'm making a diffusion model myself, but my data is not images. Now that SD3 is out (hell yeah), I'm trying to implement rectified flow to make my model better, but I am not sure about the specific scheduling process from the paper (https://arxiv.org/pdf/2403.03206). Is it one to one the same as in https://huggingface.co/docs/diffusers/en/api/schedulers/flow_match_euler_discrete ?

I can try to recreate it myself based on the paper (https://github.com/gnobitab/RectifiedFlow), but there seems to be different implementations of rectified flow, and it does not look straight forward.

Edit: I realized it cannot be the same implementation, we are missing the entire logit-Norm sampling of timesteps, which I cant see how is compatible with the discrete timesteps (default 100)