Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
[CVPR 2024 Highlight] DistriFusion: Distributed Parallel Inference for High-Resolution Diffusion Models
| Date | Stars |
|---|---|
| 2026-07-31 | 727 |
| 2026-08-02 | 727 |
| 2026-08-06 | 727 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
# DistriFusion: Distributed Parallel Inference for High-Resolution Diffusion Models ### [Paper](http://arxiv.org/abs/2402.19481) | [Project](https://hanlab.mit.edu/projects/distrifusion) | [Blog](https://hanlab.mit.edu/blog/distrifusion) | [Slides](https://www.dropbox.com/scl/fi/yv98hi2kdoh27ej4jqlbp/slides.key?rlkey=3rmfxpezqt3co5x2hgqvxv09i&st=ve4z9w6t&dl=0) | [Youtube](https://www.youtube.com/watch?v=EZX7srDDmW0&list=PL80kAHvQbh-pKRxcSS6xjds7U7Yc0gDQI&index=1) | [Poster](https://www.dropbox.com/scl/fi/labhefjwi9r01e3o9eob0/poster.pdf?rlkey=rjj1jj179enln92h8kygrftmg&st=0ddego10&dl=0) **[Dec 1, 2024]** DistriFusion is integrated in NVIDIA's [TensorRT-LLM](https://github.com/NVIDIA/TensorRT-LLM/blob/main/examples/sdxl/README.md) for distributed inference on high-resolution image generation. **[Jul 29, 2024]** DistriFusion is supported in [ColossalAI](https://github.com/hpcaitech/ColossalAI/blob/main/colossalai/inference/README.md)! **[Apr 4, 2024]** DistriFusion is selected as a **highlight** poster in CVPR 2024! **[Feb 29, 2024]** DistriFusion is accepted by CVPR 2024! Our code is publicly available!  *We introduce DistriFusion, a training-free algorithm to harness multiple GPUs to accelerate diffusion model inference without sacrificing image quality. Naïve Patch (Overview (b)) suffers from the fragmentation issue due to the lack of patch interaction. The presented examples are generated with SDXL using a 50-step Euler sampler at 1280×1920 resolution, and latency is measured on A100 GPUs.* DistriFusion: Distributed Parallel Inference for High-Resolution Diffusion Models</br> [Muyang Li](https://lmxyy.me/)\*, [Tianle Cai](https://www.tianle.website/)\*, [Jiaxin Cao](https://www.linkedin.com/in/jiaxin-cao-2166081b3/), [Qinsheng Zhang](https://qsh-zh.github.io), [Han Cai](https://han-cai.github.io), [Junjie Bai](https://www.linkedin.com/in/junjiebai/), [Yangqing Jia](https://daggerfs.com), [Ming-Yu Liu](https://mingyuliu.net), [Kai Li](https://www.cs.princeton.edu/~li/), and [Song Han](https://hanlab.mit.edu/songhan)</br> MIT, Princeton, Lepton AI, and NVIDIA</br> In CVPR 2024. ## Overview  **(a)** Original diffusion model running on a single device. **(b)** Naïvely splitting the image into 2 patches across 2 GPUs has an evident seam at the boundary due to the absence of interaction across patches. **(c)** Our DistriFusion employs synchronous communication for patch interaction at the first step. After that, we reuse the activations from the previous step via asynchronous communication. In this way, the communication overhead can be hidden into the computation pipeline. ## Performance ### Speedups <p align="center"> <img src="https://github.com/mit-han-lab/distrifuser/blob/main/assets/speedups.jpg" width="80%"/> </p>Measured total latency of DistriFusion with SDXL using a 50-step DDIM sampler for generating a single image across on NVIDIA A100 GPUs. When scaling up the resolution, the GPU devices are better utilized. Remarkably, when generating 3840×3840 images, DistriFusion achieves 1.8×, 3.4× and 6.1× speedups with 2, 4, and 8 A100s, respectively. ### Quality  Qualitative results of SDXL. FID is computed against the ground-truth images. Our DistriFusion can reduce the latency according to the number of used devices while preserving visual fidelity. References: * Denoising Diffusion Implicit Model (DDIM), Song *et al.*, ICLR 2021 * Elucidating the Design Space of Diffusion-Based Generative Models, Karras *et al.*, NeurIPS 2022 * Parallel Sampling of Diffusion Models, Shih *et al.*, NeurIPS 2023 * SDXL: Improving Latent Diffusion Models for High-Resolution Image Synthesis, Podell *et al.*, ICLR 2024 ## Prerequisites * Python3 * NVIDIA GPU + CUDA >= 12.0 and corresponding
Excerpt of 10,925 characters
Read on GitHub
Muyang Li · Massachusetts Institute of Technology · United States
31
Would you bet a product on this? Bounded 0–100 and slow moving.
matched fp:c85a4424b8ea1e3f, topic:diffusion-models