optimum-nvidia
stable-fast
| optimum-nvidia | stable-fast | |
|---|---|---|
| 1 | 11 | |
| 792 | 979 | |
| 10.3% | - | |
| 9.3 | 9.4 | |
| about 2 hours ago | 14 days ago | |
| Python | Python | |
| Apache License 2.0 | MIT License |
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
optimum-nvidia
stable-fast
-
Has anyone managed to get TensorRT working in ComfyUI on Windows?
Download (https://github.com/chengzeyi/stable-fast/releases) and install stable-fast binary, compiled according to your system: pip install stable_fast-0.0.13.post3+torch210cu118-cp310-cp310-win_amd64.whl
- Optimum-NVIDIA - 28x faster inference in just 1 line of code !?
- stable-fast for SD inference: Faster than AITemplate, On par with TensorRT
- [N] stable-fast for SD inference: Faster than AITemplate, On par with TensorRT
- Stable-fast for SD inference: Faster than AITemplate, On par with TensorRT
-
SDXL Turbo: A Real-Time Text-to-Image Generation Model
SDXL and ControlNet are already optimized, if thats what you mean: https://github.com/chengzeyi/stable-fast
(Note the links to various SD compilers).
But the whole field is moving so fast that people aren't even adopting the compilers at large.
-
Getting sub 100ms refresh rate on LCMs
> already compiling
Hmm, well if you mean torch.compile, y'all should still check out stable-fast, which is claiming ~16ms/iter on a 4090:
https://github.com/chengzeyi/stable-fast#rtx-4090-512x512-ba...
-
Generate images fast with SD 1.5 while typing on Gradio
Now combine this with an optimized SD implementation, like:
https://github.com/chengzeyi/stable-fast
Or AITemplate, and you are at 15FPS on a larger consumer GPU. 10 with a controlnet you can use for some motion consistency.
-
S-LoRA: Serving Concurrent LoRA Adapters
Since I am sending you down the rabbit hole anyway, you should check out sfast:
https://github.com/chengzeyi/stable-fast
It's, the most promising "fast" and flexible stable diffusion implementation akin to this paper or vLLM that I know of. It doesn't have as many caveats as other implementations, like AITemplate (which is basically Turing+ and linux only) or torch.compile (basically no support for changing inputs/loras).
-
🚀Announcing stable-fast v0.0.5: Speed Optimization for SDXL, Dynamic CUDA Graph
About 2 weeks ago, I released the stable-fast project, which is a lightweight inference performance optimization framework for HuggingFace Diffusers. It provides best performance while keeping the compilation dynamic and flexible, and supports ControlNet and LoRA seamlessly.
What are some alternatives?
TensorRT-LLM - TensorRT-LLM provides users with an easy-to-use Python API to define Large Language Models (LLMs) and build TensorRT engines that contain state-of-the-art optimizations to perform inference efficiently on NVIDIA GPUs. TensorRT-LLM also contains components to create Python and C++ runtimes that execute those TensorRT engines.
Fooocus - Focus on prompting and generating
gpt-fast - Simple and efficient pytorch-native transformer text generation in <1000 LOC of python.