syunar / onediff

OneDiff: An out-of-the-box acceleration library for diffusion models.

OneDiff

An out-of-the-box acceleration library for diffusion models (especially for ComfyUI, HF diffusers, and Stable Diffusion web UI).

Easy to use

• Acceleration for popular UIs/libs
  • ComfyUI
  • HF diffusers 🤗
  • Stable Diffusion web UI
• Acceleration for state-of-the-art Models
  • SDXL
  • SDXL Turbo
  • SD 1.5/2.1
  • LoRA (and dynamic switching LoRA)
  • ControlNet
  • LCM and LCM LoRA
  • Stable Video Diffusion
  • DeepCache for ComfyUI
• Out-of-the-box acceleration
  • ComfyUI Nodes
  • Acceleration with oneflow_compile
• Multi-resolution input
• Compile and save the compiled result offline, then load it online for serving
  • Save and Load
  • Change device to do multi-process serving

State-of-the-art performance

Updated on Nov 6, 2023.

Device	SD1.5 (512x512)	SD2.1 (512x512)	SDXL1.0-base（1024x1024）
RTX 3090	42.38it/s	42.33it/s	6.66it/s
RTX 4090	74.71it/s	73.57it/s	13.57it/s
A100-PCIE-40GB	54.4it/s	54.06it/s	10.22it/s
A100-SXM4-80GB	59.68it/s	61.91it/s	11.80it/s

NOTE: OneDiff Enterprise Edition delivers even higher performance and second-to-none deployment flexibility.

OS and GPU support

• Linux
  • If you want to use OneDiff on Windows, please use it under WSL.
• NVIDIA GPUs

Need help or talk

• Discord of OneDiff
• GitHub issues

OneDiff Enterprise Edition

If you need Enterprise Level Support for your system or business, please send an email to [e-mail] and tell us about your user case, deployment scale, and requirements.

OneDiff Enterprise Edition can be subscripted for one month and one GPU and the cost is low: https://siliconflow.com/product.html

	OneDiff Enterprise	OneDiff Community
SD/SDXL series model Optimization	Yes	Yes
UNet/VAE/ControlNet Optimization	Yes	Yes
LoRA(and dynamic switching LoRA)	Yes	Yes
SDXL Turbo/LCM	Yes	Yes
Stable Video Diffusion	Yes	Yes
HF diffusers	Yes	Yes
ComfyUI	Yes	Yes
Stable Diffusion web UI	Yes	Yes
Multiple Resolutions	Yes(No time cost for most of the cases)	Yes(Cost a few seconds/minutes to compile for new input shape)
Technical Support for deployment	High priority support	Community
More Extreme and Dedicated optimization(usually another 20~50% performance gain)	Yes
Support customized pipeline/workflow	Yes
Get the latest technology/feature	Yes

Install from source or Using in Docker

Install from source

1. Install OneFlow

NOTE: We have updated OneFlow a lot for OneDiff, so please install OneFlow by the links below.

For CUDA 11.8

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

Click to get OneFlow packages for other CUDA versions.

CUDA 12.1

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

CUDA 12.2

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

2. Install torch and diffusers

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

3. Install OneDiff

git clone https://github.com/siliconflow/onediff.git
cd onediff && python3 -m pip install -e .

4. (Optional)Login huggingface-cli

python3 -m pip install huggingface_hub
 ~/.local/bin/huggingface-cli login

Docker

docker pull oneflowinc/onediff:20231106