Which formulation used in ddpm_sample() implementation?

Hi,

thanks for the implementation!

I am trying to understand how you implemented method ddpm_sample, in particular, how you incorporated "time_next" into into the denoising step. Which paper did you use as guidance? I was comparing your ddpm_sample with Algorithm 2 of paper "Denoising diffusion probabilistic models" but can't match your use of expm1, for instance.

Thanks a lot.

Start at the ddpm_sample() implementation and trace how time_next and expm1 are used in the denoising step. Compare that flow with Algorithm 2 of the cited Denoising diffusion probabilistic models paper; the work is done when the formulation and paper or derivation guiding it are clearly explained.

Written by the indexing model from the issue text.