change of beta_schedule leads to significantly worse results

[BillyChern]

I just altered the beta_schedule parameter in the class GaussianDiffusion from the default 'sigmoid' to 'linear', which was utilized in the original ddpm paper. However, the results were unreasonably much worse than the 'sigmoid' edition. Actually, it was a mess. I can't figure out why. Has anyone encountered similar kinds of issues? It worked well with the 'sigmoid' schedule.