A PyTorch-native inference engine with cache, parallelism, quantization and cpu offload for 100+ DiTs.

cache-dit

该项目是为 Diffusers 提供统一缓存加速的框架，支持几乎所有的 DiT 扩散模型，包括 Qwen-Image-Lightning、Qwen-Image、HunyuanImage、Wan、FLUX 等。它通过简单的代码即可实现高效的缓存加速功能，无需重新训练模型即可显著提升推理速度。

This project is a framework that provides unified caching acceleration for Diffusers. It supports almost all DiT diffusion models, including Qwen-Image-Lightning, Qwen-Image, HunyuanImage, Wan, FLUX, etc. It can achieve efficient caching acceleration through simple code and significantly improve the inference speed without retraining the model.

日期	Star	较前日
周日	1,297	—
周一	1,298	+1
周二	1,298	0
周三	1,298	0
周四	1,300	+2
周五	1,301	+1
周六	1,302	+1

vipshop/cache-dit

vipshop/cache-dit

评论