Skip to content


Repository files navigation

Scalable Diffusion Models with Transformers (DiT)
Official PyTorch Implementation

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

DiT samples

This repo contains PyTorch model definitions, pre-trained weights and training/sampling code 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:

An implementation of DiT directly in Hugging Face diffusers can also be found here.


First, download and set up the repo:

git clone
cd DiT

We provide an environment.yml file that can be used to create a Conda environment. If you only want to run pre-trained 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 Hugging Face Spaces Open In Colab

More DiT samples

Pre-trained DiT checkpoints. You can sample from our pre-trained DiT models with Weights for our pre-trained DiT model will be automatically downloaded depending on the model you use. The script has various arguments to switch between the 256x256 and 512x512 models, adjust sampling steps, change the classifier-free guidance scale, etc. For example, to sample from our 512x512 DiT-XL/2 model, you can use:

python --image-size 512 --seed 1

For convenience, our pre-trained 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

Custom DiT checkpoints. If you've trained a new DiT model with (see below), you can add the --ckpt argument to use your own checkpoint instead. For example, to sample from the EMA weights of a custom 256x256 DiT-L/4 model, run:

python --model DiT-L/4 --image-size 256 --ckpt /path/to/

Training DiT

We provide a training script for DiT in This script can be used to train class-conditional DiT models, but it can be easily modified to support other types of conditioning. To launch DiT-XL/2 (256x256) training with N GPUs on one node:

torchrun --nnodes=1 --nproc_per_node=N --model DiT-XL/2 --data-path /path/to/imagenet/train

PyTorch Training Results

We've trained DiT-XL/2 and DiT-B/4 models from scratch with the PyTorch training script to verify that it reproduces the original JAX results up to several hundred thousand training iterations. Across our experiments, the PyTorch-trained models give similar (and sometimes slightly better) results compared to the JAX-trained models up to reasonable random variation. Some data points:

DiT Model Train Steps FID-50K
(JAX Training)
(PyTorch Training)
PyTorch Global Training Seed
XL/2 400K 19.5 18.1 42
B/4 400K 68.4 68.9 42
B/4 400K 68.4 68.3 100

These models were trained at 256x256 resolution; we used 8x A100s to train XL/2 and 4x A100s to train B/4. Note that FID here is computed with 250 DDPM sampling steps, with the mse VAE decoder and without guidance (cfg-scale=1).

TF32 Note (important for A100 users). When we ran the above tests, TF32 matmuls were disabled per PyTorch's defaults. We've enabled them at the top of and because it makes training and sampling way way way faster on A100s (and should for other Ampere GPUs too), but note that the use of TF32 may lead to some differences compared to the above results.


Training (and sampling) could likely be sped-up significantly by:

  • using Flash Attention in the DiT model
  • using torch.compile in PyTorch 2.0

Basic features that would be nice to add:

  • Monitor FID and other metrics
  • Generate and save samples from the EMA model periodically
  • Resume training from a checkpoint
  • AMP/bfloat16 support

🔥 Feature Update Check out this repository at to preview a selection of training speed acceleration and memory saving features including gradient checkpointing, mixed precision training and pre-extrated VAE features. With these advancements, we have achieved a training speed of 0.84 steps/sec for DiT-XL/2 using just a single A100 GPU.

Evaluation (FID, Inception Score, etc.)

We include a script which samples a large number of images from a DiT model in parallel. This script generates a folder of samples as well as a .npz file which can be directly used with ADM's TensorFlow evaluation suite to compute FID, Inception Score and other metrics. For example, to sample 50K images from our pre-trained DiT-XL/2 model over N GPUs, run:

torchrun --nnodes=1 --nproc_per_node=N --model DiT-XL/2 --num-fid-samples 50000

There are several additional options; see for details.

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).


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


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.


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