autoregressive-diffusion-pytorch 0.4.0

Autoregressive Diffusion - Pytorch

Implementation of the architecture behind Autoregressive Image Generation without Vector Quantization in Pytorch

Official repository has been released here

Alternative route

oxford flowers at 96k steps

Install

$ pip install autoregressive-diffusion-pytorch

Usage

import torch
from autoregressive_diffusion_pytorch import AutoregressiveDiffusion

model = AutoregressiveDiffusion(
    dim_input = 512,
    dim = 1024,
    max_seq_len = 32,
    depth = 8,
    mlp_depth = 3,
    mlp_width = 1024
)

seq = torch.randn(3, 32, 512)

loss = model(seq)
loss.backward()

sampled = model.sample(batch_size = 3)

assert sampled.shape == seq.shape

For images treated as a sequence of tokens (as in paper)

import torch
from autoregressive_diffusion_pytorch import ImageAutoregressiveDiffusion

model = ImageAutoregressiveDiffusion(
    model = dict(
        dim = 1024,
        depth = 12,
        heads = 12,
    ),
    image_size = 64,
    patch_size = 8
)

images = torch.randn(3, 3, 64, 64)

loss = model(images)
loss.backward()

sampled = model.sample(batch_size = 3)

assert sampled.shape == images.shape

An images trainer

import torch

from autoregressive_diffusion_pytorch import (
    ImageDataset,
    ImageAutoregressiveDiffusion,
    ImageTrainer
)

dataset = ImageDataset(
    '/path/to/your/images',
    image_size = 128
)

model = ImageAutoregressiveDiffusion(
    model = dict(
        dim = 512
    ),
    image_size = 128,
    patch_size = 16
)

trainer = ImageTrainer(
    model = model,
    dataset = dataset
)

trainer()

For an improvised version using flow matching, just import ImageAutoregressiveFlow and AutoregressiveFlow instead

The rest is the same

ex.

import torch

from autoregressive_diffusion_pytorch import (
    ImageDataset,
    ImageTrainer,
    ImageAutoregressiveFlow,
)

dataset = ImageDataset(
    '/path/to/your/images',
    image_size = 128
)

model = ImageAutoregressiveFlow(
    model = dict(
        dim = 512
    ),
    image_size = 128,
    patch_size = 16,
    xm_candidates = 4  # explorative modeling - train against best of k candidates per timestep
)

trainer = ImageTrainer(
    model = model,
    dataset = dataset
)

trainer()

Citations

@article{Li2024AutoregressiveIG,
    title   = {Autoregressive Image Generation without Vector Quantization},
    author  = {Tianhong Li and Yonglong Tian and He Li and Mingyang Deng and Kaiming He},
    journal = {ArXiv},
    year    = {2024},
    volume  = {abs/2406.11838},
    url     = {https://api.semanticscholar.org/CorpusID:270560593}
}

@article{Wu2023ARDiffusionAD,
    title     = {AR-Diffusion: Auto-Regressive Diffusion Model for Text Generation},
    author    = {Tong Wu and Zhihao Fan and Xiao Liu and Yeyun Gong and Yelong Shen and Jian Jiao and Haitao Zheng and Juntao Li and Zhongyu Wei and Jian Guo and Nan Duan and Weizhu Chen},
    journal   = {ArXiv},
    year      = {2023},
    volume    = {abs/2305.09515},
    url       = {https://api.semanticscholar.org/CorpusID:258714669}
}

@article{Karras2022ElucidatingTD,
    title   = {Elucidating the Design Space of Diffusion-Based Generative Models},
    author  = {Tero Karras and Miika Aittala and Timo Aila and Samuli Laine},
    journal = {ArXiv},
    year    = {2022},
    volume  = {abs/2206.00364},
    url     = {https://api.semanticscholar.org/CorpusID:249240415}
}

@article{Liu2022FlowSA,
    title   = {Flow Straight and Fast: Learning to Generate and Transfer Data with Rectified Flow},
    author  = {Xingchao Liu and Chengyue Gong and Qiang Liu},
    journal = {ArXiv},
    year    = {2022},
    volume  = {abs/2209.03003},
    url     = {https://api.semanticscholar.org/CorpusID:252111177}
}

@article{Esser2024ScalingRF,
    title   = {Scaling Rectified Flow Transformers for High-Resolution Image Synthesis},
    author  = {Patrick Esser and Sumith Kulal and A. Blattmann and Rahim Entezari and Jonas Muller and Harry Saini and Yam Levi and Dominik Lorenz and Axel Sauer and Frederic Boesel and Dustin Podell and Tim Dockhorn and Zion English and Kyle Lacey and Alex Goodwin and Yannik Marek and Robin Rombach},
    journal = {ArXiv},
    year    = {2024},
    volume  = {abs/2403.03206},
    url     = {https://api.semanticscholar.org/CorpusID:268247980}
}

@misc{li2025basicsletdenoisinggenerative,
    title   = {Back to Basics: Let Denoising Generative Models Denoise},
    author  = {Tianhong Li and Kaiming He},
    year    = {2025},
    eprint  = {2511.13720},
    archivePrefix = {arXiv},
    primaryClass = {cs.CV},
    url     = {https://arxiv.org/abs/2511.13720},
}

@misc{gladstone2026explorativemodelingunlockingpretraining,
    title   = {Explorative Modeling: Unlocking a Third Pretraining Axis and End-to-End Generation},
    author  = {Alexi Gladstone and Heng Ji and Yilun Du},
    year    = {2026},
    eprint  = {2607.27372},
    archivePrefix = {arXiv},
    primaryClass = {cs.LG},
    url     = {https://arxiv.org/abs/2607.27372},
}