---
license: bigscience-openrail-m
datasets:
- ILSVRC/imagenet-1k
---
# SmallDiT
复现经典的DiT工作([Scalable Diffusion Models with Transformers](https://arxiv.org/abs/2212.09748))，训练数据为ImageNet.
代码仓库: `https://github.com/lixiang90/ClassicalModels`
## vae
`vae.pt`是用于图像压缩的vae模型，把(256,256,3)的图像压缩为(32,32,4)的latents.