参考：
https://github.com/huggingface/blog/blob/main/quanto-diffusers.md

安装

pip install optimum-quanto
%pip install optimum

使用

from optimum.quanto import freeze, qfloat8, quantize
from diffusers import PixArtSigmaPipeline
import torch

pipeline = PixArtSigmaPipeline.from_pretrained