Awesome
<p align="center"> <img src=".github/stochastic_logo_light.svg#gh-light-mode-only" width="250" alt="Stochastic.ai"/> <img src=".github/stochastic_logo_dark.svg#gh-dark-mode-only" width="250" alt="Stochastic.ai"/> </p> <br> <!-- # ⚡️ Real-time inference for Stable Diffusion --> <!--  --> <!-- <p align="center"> <img src=".github/stochasticai_demo.gif" width="600" alt="Stochastic.ai Demo"/> </p> -->Welcome to x-stable-diffusion by Stochastic!
This project is a compilation of acceleration techniques for the Stable Diffusion model to help you generate images faster and more efficiently, saving you both time and money.
With example images and a comprehensive benchmark, you can easily choose the best technique for your needs. When you're ready to deploy, our CLI called stochasticx makes it easy to get started on your local machine. Try x-stable-diffusion and see the difference it can make for your image generation performance and cost savings.
🚀 Installation
Quickstart
Make sure you have Python and Docker installed on your system
- Install the latest version of
stochasticxlibrary.
pip install stochasticx
- Deploy the Stable Diffusion model
stochasticx stable-diffusion deploy --type aitemplate
Alternatively, you can deploy stable diffusion without our CLI by checking the steps here.
- To perform inference with this deployed model:
stochasticx stable-diffusion inference --prompt "Riding a horse"
Check all the options of the inference command:
stochasticx stable-diffusion inference --help
- You can get the logs of the deployment executing the following command:
stochasticx stable-diffusion logs
- Stop and remove the deployment with this command:
stochasticx stable-diffusion stop
How to get less than 1s latency?
Change the num_inference_steps to 30. With this, you can get an image generated in 0.88 seconds.
{
'max_seq_length': 64,
'num_inference_steps': 30,
'image_size': (512, 512)
}
You can also experiment with reducing the image_size.
How to run on Google Colab?
In each folder, we will provide a Google Colab notebook with which you can test the full flow and inference on a T4 GPU
Manual deployment
Check the README.md of the following directories:
🔥 Optimizations
- AITemplate: Latest optimization framework of Meta
- TensorRT: NVIDIA TensorRT framework
- nvFuser: nvFuser with Pytorch
- FlashAttention: FlashAttention intergration in Xformers
Benchmarks
Setup
For hardware, we used 1x40GB A100 GPU with CUDA 11.6 and the results are reported by averaging 50 runs.
The following arguments were used for image generation for all the benchmarks:
{
'max_seq_length': 64,
'num_inference_steps': 50,
'image_size': (512, 512)
}
Online results
For batch_size 1, these are the latency results:
A100 GPU
| project | Latency (s) | GPU VRAM (GB) |
|---|---|---|
| PyTorch fp16 | 5.77 | 10.3 |
| nvFuser fp16 | 3.15 | --- |
| FlashAttention fp16 | 2.80 | 7.5 |
| TensorRT fp16 | 1.68 | 8.1 |
| AITemplate fp16 | 1.38 | 4.83 |
| ONNX (CUDA) | 7.26 | 13.3 |
T4 GPU
Note: AITemplate might not support T4 GPU yet. Check support here
| project | Latency (s) |
|---|---|
| PyTorch fp16 | 16.2 |
| nvFuser fp16 | 19.3 |
| FlashAttention fp16 | 13.7 |
| TensorRT fp16 | 9.3 |
Batched results - A100 GPU
The following results were obtained by varying batch_size from 1 to 24.
| project \ bs | 1 | 4 | 8 | 16 | 24 |
|---|---|---|---|---|---|
| Pytorch fp16 | 5.77s/10.3GB | 19.2s/18.5GB | 36s/26.7GB | OOM | |
| FlashAttention fp16 | 2.80s/7.5GB | 9.1s/17GB | 17.7s/29.5GB | OOM | |
| TensorRT fp16 | 1.68s/8.1GB | OOM | |||
| AITemplate fp16 | 1.38s/4.83GB | 4.25s/8.5GB | 7.4s/14.5GB | 15.7s/25GB | 23.4s/36GB |
| ONNX (CUDA) | 7.26s/13.3GB | OOM | OOM | OOM | OOM |
Note: TensorRT fails to convert UNet model from ONNX to TensorRT due to memory issues.
Sample images generated
Click here to view the complete list of generated images
| Optimization \ Prompt | Super Mario learning to fly in an airport, Painting by Leonardo Da Vinci | The Easter bunny riding a motorcycle in New York City | Drone flythrough of a tropical jungle convered in snow |
|---|---|---|---|
| PyTorch fp16 |  |  |  |
| nvFuser fp16 |  |  |  |
| FlashAttention fp16 |  |  |  |
| TensorRT fp16 |  |  |  |
| AITemplate fp16 |  |  |  |
References
🌎 Join our community
- Discord - https://discord.gg/TgHXuSJEk6
🌎 Contributing
As an open source project in a rapidly evolving field, we welcome contributions of all kinds, including new features and better documentation. Please read our contributing guide to learn how you can get involved.
<!-- ## Team and contributors `x-stable-diffusion` is a community-driven project with several AI systems engineers and researchers contributing to it. It is currently maintained by: [Toan Do](https://github.com/Toan-Do), [Marcos Rivera](https://github.com/MarcosRiveraMartinez), [Sarthak Langde](https://github.com/sarthaklangde), [Subhash GN](https://github.com/subhash-stc), [Riccardo Romagnoli](https://github.com/RiccardoRomagnoli), [Roman Ageev](https://github.com/StochasticRomanAgeev) and [Glenn Ko](https://github.com/glennko) --> <!-- ## ✅ Stochastic Stochastic was founded with a vision to make deep learning optimization and deployment effortless. With our cloud platform, you can easily optimize and deploy your deep learning models with confidence, knowing that you are getting the best performance possible. Our platform automatically optimizes your models, benchmarking them on various evaluation metrics to ensure they are running at their peak. And when it comes time to deploy, Stochastic has you covered with auto-scaling accelerated inference for models like BLOOM 176B, Stable Diffusion, and GPT-J. Plus, our platform is cloud agnostic, supporting AWS, GCP, Azure, and Kubernetes clusters <p align="center"> <img src=".github/stochastic_x_dashboard.jpeg" width="600" alt="Stochastic X Dashboard"/> </p> --> <!-- For fully-managed solution hosted on Stochastic [Sign up →](https://app.stochastic.ai/signup) --> </br > For managed hosting on our cloud or on your private cloud [Contact us →](https://stochastic.ai/contact)