Awesome

onediff is an out-of-the-box acceleration library for diffusion models, it provides:

• Out-of-the-box acceleration for popular UIs/libs(such as HF diffusers and ComfyUI)
• PyTorch code compilation tools and strong optimized GPU Kernels for diffusion models

News

• [2024/07/23] :rocket: Up to 1.7x Speedup for Kolors: Kolors Acceleration Report
• [2024/06/18] :rocket: Acceleration for DiT models: SD3 Acceleration Report, PixArt Acceleration Report, and Latte Acceleration Report
• [2024/04/13] :rocket: OneDiff 1.0 is released (Acceleration of SD & SVD with one line of code)
• [2024/01/12] :rocket: Accelerating Stable Video Diffusion 3x faster with OneDiff DeepCache + Int8
• [2023/12/19] :rocket: Accelerating SDXL 3x faster with DeepCache and OneDiff

Hiring

We're hiring! If you are interested in working on onediff at SiliconFlow, we have roles open for Interns and Engineers in Beijing (near Tsinghua University).

If you have contributed significantly to open-source software and are interested in remote work, you can contact us at talent@siliconflow.cn with onediff in the email title.

• Documentation
  • Use with HF diffusers and ComfyUI
  • Performance comparison
    • SDXL E2E time
    • SVD E2E time
  • Quality Evaluation
  • Community and Support
  • Installation
    • 0. OS and GPU Compatibility
    • 1. Install torch and diffusers
    • 2. Install a compiler backend
      • Nexfort
      • OneFlow
    • 3. Install onediff
• More about onediff
  • Architecture
  • Features
  • Acceleration for State-of-the-art models
  • Acceleration for production environment
    • PyTorch Module compilation
    • Avoid compilation time for new input shape
    • Avoid compilation time for online serving
    • Distributed Run
  • OneDiff Enterprise Solution

Documentation

onediff is the abbreviation of "one line of code to accelerate diffusion models".

Use with HF diffusers and ComfyUI

• onediff for HF diffusers 🤗
• onediff for ComfyUI
• onediff for Stable Diffusion web UI

Performance comparison

SDXL E2E time

• Model stabilityai/stable-diffusion-xl-base-1.0;
• Image size 1024*1024, batch size 1, steps 30;
• NVIDIA A100 80G SXM4;

SVD E2E time

• Model stabilityai/stable-video-diffusion-img2vid-xt;
• Image size 576*1024, batch size 1, steps 25, decoder chunk size 5;
• NVIDIA A100 80G SXM4;

Note that we haven't got a way to run SVD with TensorRT on Feb 29 2024.

Quality Evaluation

We also maintain a repository for benchmarking the quality of generation after acceleration: odeval

Community and Support

• Create an issue
• Chat in Discord:
• Community and Feedback

Installation

0. OS and GPU Compatibility

• Linux
  • If you want to use onediff on Windows, please use it under WSL.
  • The guide to install onediff in WSL2.
• NVIDIA GPUs
  • Compatibility with Nvidia GPUs.

1. Install torch and diffusers

Note: You can choose the latest versions you want for diffusers or transformers.

python3 -m pip install "torch" "transformers==4.27.1" "diffusers[torch]==0.19.3"

2. Install a compiler backend

When considering the choice between OneFlow and Nexfort, either one is optional, and only one is needed.

• For DiT structural models or H100 devices, it is recommended to use Nexfort.

• For all other cases, it is recommended to use OneFlow. Note that optimizations within OneFlow will gradually transition to Nexfort in the future.

Nexfort

Install Nexfort is Optional. The detailed introduction of Nexfort is here.

python3 -m  pip install -U torch==2.3.0 torchvision==0.18.0 torchaudio==2.3.0 torchao==0.1
python3 -m  pip install -U nexfort

OneFlow

Install OneFlow is Optional.

NOTE: We have updated OneFlow frequently for onediff, so please install OneFlow by the links below.

• CUDA 11.8

  For NA/EU users

  python3 -m pip install -U --pre oneflow -f https://github.com/siliconflow/oneflow_releases/releases/expanded_assets/community_cu118
  

  For CN users

  python3 -m pip install -U --pre oneflow -f https://oneflow-pro.oss-cn-beijing.aliyuncs.com/branch/community/cu118
  

• CUDA 12.1

  For NA/EU users

  python3 -m pip install -U --pre oneflow -f https://github.com/siliconflow/oneflow_releases/releases/expanded_assets/community_cu121
  

  For CN users

  python3 -m pip install -U --pre oneflow -f https://oneflow-pro.oss-cn-beijing.aliyuncs.com/branch/community/cu121
  

• CUDA 12.2

  For NA/EU users

  python3 -m pip install -U --pre oneflow -f https://github.com/siliconflow/oneflow_releases/releases/expanded_assets/community_cu122
  

  For CN users

  python3 -m pip install -U --pre oneflow -f https://oneflow-pro.oss-cn-beijing.aliyuncs.com/branch/community/cu122
  

3. Install onediff

• From PyPI

python3 -m pip install --pre onediff

• From source

git clone https://github.com/siliconflow/onediff.git

cd onediff && python3 -m pip install -e .

Or install for development:

# install for dev
cd onediff && python3 -m pip install -e '.[dev]'

# code formatting and linting
pip3 install pre-commit
pre-commit install
pre-commit run --all-files

NOTE: If you intend to utilize plugins for ComfyUI/StableDiffusion-WebUI, we highly recommend installing OneDiff from the source rather than PyPI. This is necessary as you'll need to manually copy (or create a soft link) for the relevant code into the extension folder of these UIs/Libs.

More about onediff

Architecture

Features

Acceleration for State-of-the-art models

onediff supports the acceleration for SOTA models.

• stable: release for public usage, and has long-term support;
• beta: release for professional usage, and has long-term support;
• alpha: early release for expert usage, and should be careful to use;

Acceleration for production environment

PyTorch Module compilation

• compilation with oneflow_compile

Avoid compilation time for new input shape

• Support Multi-resolution input

Avoid compilation time for online serving

Compile and save the compiled result offline, then load it online for serving

• Save and Load the compiled graph
• Compile at one device(such as device 0), then use the compiled result to other device(such as device 1~7). Change device of the compiled graph to do multi-process serving

Distributed Run

If you want to do distributed inference, you can use onediff's compiler to do single-device acceleration in a distributed inference engine such as xDiT

OneDiff Enterprise Solution

If you need Enterprise-level Support for your system or business, you can email us at contact@siliconflow.com, or contact us through the website: https://siliconflow.cn/pricing

Citation

@misc{2022onediff,
  author={OneDiff Contributors},
  title = {OneDiff: An out-of-the-box acceleration library for diffusion models},
  year = {2022},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/siliconflow/onediff}}
}