/DiT

Official PyTorch Implementation of "Scalable Diffusion Models with Transformers"

Primary LanguagePythonOtherNOASSERTION

Scalable Diffusion Models with Transformers (DiT)
Official PyTorch Implementation

Paper | Project Page | Run DiT-XL/2 Open In Colab

DiT samples

This repo contains PyTorch model definitions, sampling code and pre-trained weights for our paper exploring diffusion models with transformers (DiTs). You can find more visualizations on our project page.

Scalable Diffusion Models with Transformers
William Peebles, Saining Xie
UC Berkeley, New York University

We train latent diffusion models, replacing the commonly-used U-Net backbone with a transformer that operates on latent patches. We analyze the scalability of our Diffusion Transformers (DiTs) through the lens of forward pass complexity as measured by Gflops. We find that DiTs with higher Gflops---through increased transformer depth/width or increased number of input tokens---consistently have lower FID. In addition to good scalability properties, our DiT-XL/2 models outperform all prior diffusion models on the class-conditional ImageNet 512×512 and 256×256 benchmarks, achieving a state-of-the-art FID of 2.27 on the latter.

This repository contains:

  • 🪐 A simple PyTorch implementation of DiT
  • ⚡️ Pre-trained class-conditional DiT models trained on ImageNet (512x512 and 256x256)
  • 💥 A self-contained Colab notebook for running pre-trained DiT-XL/2 models

Setup

First, download and set up the repo:

git clone https://github.com/facebookresearch/DiT.git
cd DiT

We provide an environment.yml file that can be used to create a Conda environment. If you only want to run the models locally on CPU, you can remove the cudatoolkit and pytorch-cuda requirements from the file.

conda env create -f environment.yml
conda activate DiT

Sampling Open In Colab

More DiT samples

You can sample from our pre-trained DiT models with main.py. Weights for our pre-trained DiT model will be automatically downloaded depending on the model you use (default: 256x256 XL/2 model). You can edit the script to use the 512x512 model, adjust sampling steps, input classes and classifier-free guidance scale, etc.

For convenience, the DiT models can be downloaded directly here as well:

DiT Model Image Resolution FID-50K Inception Score Gflops
XL/2 256x256 2.27 278.24 119
XL/2 512x512 3.04 240.82 525

Differences from JAX

Our models were originally trained in JAX on TPUs. The weights in this repo are ported directly from the JAX models. There may be minor differences in results stemming from sampling with different floating point precisions. We re-evaluated our ported PyTorch weights at FP32, and they actually perform marginally better than sampling in JAX (2.21 FID versus 2.27 in the paper).

BibTeX

@article{Peebles2022DiT,
  title={Scalable Diffusion Models with Transformers},
  author={William Peebles and Saining Xie},
  year={2022},
  journal={arXiv preprint arXiv:2212.09748},
}

Acknowledgments

We thank Kaiming He, Ronghang Hu, Alexander Berg, Shoubhik Debnath, Tim Brooks, Ilija Radosavovic and Tete Xiao for helpful discussions. William Peebles is supported by the NSF Graduate Research Fellowship.

This codebase borrows from OpenAI's diffusion repos, most notably ADM.

License

The code and model weights are licensed under CC-BY-NC. See LICENSE.txt for details.