[feature request] Support for training model with v prediction

Not sure I'm requesting the proper things from this paper... but would love to see the "fixes" from the below paper available as training options:

https://arxiv.org/abs/2305.08891

Relevant Reddit thread: https://www.reddit.com/r/StableDiffusion/comments/13joe98/sds_noise_schedule_is_flawed_this_new_paper/

Start by reading the linked paper (arXiv:2305.08891) and the Reddit discussion to determine which v-prediction training changes are requested. The issue does not name files, tests, or an entry point; the work is done when the identified changes are available as explicit training options and are covered by appropriate validation.

Written by the indexing model from the issue text.