[Schedulers] Add exponential sigmas / exponential noise schedule

[hlky]

What does this PR do?

This PR adds exponential sigmas. Added only to Euler for now, after initial review we will add it to other schedulers, and add the other schedulers in the linked issue in subsequent PRs.

The implementation is quite simple, we add use_exponential_sigmas parameter and _convert_to_exponential function.

Example usage:

from diffusers import StableDiffusionXLPipeline, EulerDiscreteScheduler
import torch

pipeline: StableDiffusionXLPipeline = StableDiffusionXLPipeline.from_pretrained(
    "stabilityai/stable-diffusion-xl-base-1.0",
    variant="fp16",
    torch_dtype=torch.float16,
)
pipeline.scheduler = EulerDiscreteScheduler.from_config(
    pipeline.scheduler.config, timestep_spacing="linspace", use_exponential_sigmas=True
)

Note that we set timestep_spacing="linspace" in addition to use_exponential_sigmas=True, this is because XL's scheduler config uses timestep_spacing="leading" by default and the intent is to match results from the webuis where these noise schedules are currently used. We may wish to override timestep_spacing to linspace when use_exponential_sigmas=True or add a warning.

Tested against Forge (note that cuda device for generator is essential to match Forge results):

generator = torch.Generator("cuda").manual_seed(49136503742430)

image = pipeline(
    prompt="enormous kirby. space background",
    num_inference_steps=20,
    guidance_scale=5.0,
    generator=generator,
).images[0]
image

and with the same settings on Forge we get the same result:

For reference with the default XL scheduler in Diffusers we get:

#9490

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.
cc @asomoza @yiyixuxu

[yiyixuxu]

thanks!

[yiyixuxu]

need a make style now

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[yiyixuxu]

thanks!

[yiyixuxu]

thanks for the PR! this is cool and feel free to add to other schedulers!