sample_dpmpp_2m slightly different from paper

[patrickvonplaten]

When debugging the sample_dpmpp_2m sampler I noticed that the timesteps that are passed to the model are slightly different from the paper.

More precisely, the final timestep in k-diffusion is always t=0.0 where as the paper states in Appendix B.1 (see paper here: https://arxiv.org/pdf/2211.01095.pdf):

In practice, these discrete-time DPMs usually choose uniform time steps between [0, T], thus tn =nT/N
, for n = 1, . . . , N. The smallest time is T/N.

which means the smallest timestep should not be 0.

Maybe cc @crowsonkb :-)

[patrickvonplaten]

Maybe also cc @LuChengTHU does it matter a lot whether the timesteps finish at 0.0 or not?