add the UniPC scheduler

[wl-zhao]

Add a multistep UniPC. UniPC is a training-free framework designed for the fast sampling of diffusion models, which consists of a corrector (UniC) and a predictor (UniP) that share a unified analytical form and support arbitrary orders.

For more details about the method, please refer to the [paper] and the [code].

This implementation also supports passing another scheduler as the solver_p to UniPC, which will instead perform the sampling through solver_p + UniC.

#2311

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[patrickvonplaten]

Very cool the scheduler works very well:

#!/usr/bin/env python3
from diffusers import StableDiffusionPipeline, UniPCMultistepScheduler
import time
import torch
import sys

path = "CompVis/stable-diffusion-v1-4"

pipe = StableDiffusionPipeline.from_pretrained(path, torch_dtype=torch.float16)
pipe.scheduler = UniPCMultistepScheduler.from_config(pipe.scheduler.config)
pipe = pipe.to("cuda")

prompt = "an image of Santa Claus in light blue"
generator = torch.Generator(device="cuda").manual_seed(0)
image = pipe(prompt, generator=generator, num_inference_steps=25).images[0]