[ad_1]
Addressing compatibility points throughout set up | ONNX for NVIDIA GPUs | Hugging Face’s Optimum library
This text discusses the ONNX runtime, one of the crucial efficient methods of rushing up Steady Diffusion inference. On an A100 GPU, operating SDXL for 30 denoising steps to generate a 1024 x 1024 picture may be as quick as 2 seconds. Nevertheless, the ONNX runtime is dependent upon a number of shifting items, and putting in the best variations of all of its dependencies may be difficult in a continually evolving ecosystem. Take this as a high-level debugging information, the place I share my struggles in hopes of saving you time. Whereas the precise variations and instructions may shortly turn into out of date, the high-level ideas ought to stay related for an extended time frame.
ONNX can really refer to 2 totally different (however associated) elements of the ML stack:
- ONNX is a format for storing machine studying fashions. It stands for Open Neural Community Trade and, as its identify suggests, its predominant aim is interoperability throughout platforms. ONNX is a self-contained format: it shops each the mannequin weights and structure. Because of this a single .onnx file incorporates all the data wanted to run inference. No want to jot down any further code to outline or load a mannequin; as an alternative, you merely go it to a runtime (extra on this beneath).
- ONNX can be a runtime to run mannequin which might be in ONNX format. It actually runs the mannequin. You possibly can see it as a mediator between the architecture-agnostic ONNX format and the precise {hardware} that runs inference. There’s a separate model of the runtime for every supported accelerator kind (see full listing right here). Word, nevertheless, that the ONNX runtime shouldn’t be the one strategy to run inference with a mannequin that’s in ONNX format — it’s only one method. Producers can select to construct their very own runtimes which might be hyper-optimized for his or her {hardware}. As an example, NVIDIA’s TensorRT is an alternative choice to the ONNX runtime.
This text focuses on operating Steady Diffusion fashions utilizing the ONNX runtime. Whereas the high-level ideas are in all probability timeless, be aware that the ML tooling ecosystem is in fixed change, so the precise workflow or code snippets may turn into out of date (this text was written in Might 2024). I’ll give attention to the Python implementation particularly, however be aware that the ONNX runtime also can function in different languages like C++, C#, Java or JavaScript.
Execs of the ONNX Runtime
- Stability between inference pace and interoperability. Whereas the ONNX runtime won’t all the time be the quickest answer for all sorts of {hardware}, it’s a quick sufficient answer for most varieties of {hardware}. That is significantly interesting in case you’re serving your fashions on a heterogeneous fleet of machines and don’t have the assets to micro-optimize for every totally different accelerator.
- Vast adoption and dependable authorship. ONNX was open-sourced by Microsoft, who’re nonetheless sustaining it. It’s broadly adopted and effectively built-in into the broader ML ecosystem. As an example, Hugging Face’s Optimum library lets you outline and run ONNX mannequin pipelines with a syntax that’s harking back to their widespread transformers and diffusers libraries.
Cons of the ONNX Runtime
- Engineering overhead. In comparison with the choice of operating inference immediately in PyTorch, the ONNX runtime requires compiling your mannequin to the ONNX format (which may take 20–half-hour for a Steady Diffusion mannequin) and putting in the runtime itself.
- Restricted set of ops. The ONNX format doesn’t assist all PyTorch operations (it’s much more restrictive than TorchScript). In case your mannequin is utilizing an unsupported operation, you’ll both should reimplement the related portion, or drop ONNX altogether.
- Brittle set up and setup. Because the ONNX runtime makes the interpretation from the ONNX format to architecture-specific directions, it may be difficult to get the best mixture of software program variations to make it work. As an example, if operating on an NVIDIA GPU, you want to guarantee compatibility of (1) working system, (2) CUDA model, (3) cuDNN model, and (4) ONNX runtime model. There are helpful assets just like the CUDA compatibility matrix, however you may nonetheless find yourself losing hours discovering the magic mixture that works at a given time limit.
- {Hardware} limitations. Whereas the ONNX runtime can run on many architectures, it can’t run on all architectures like pure PyTorch fashions can. As an example, there may be at the moment (Might 2024) no assist for Google Cloud TPUs or AWS Inferentia chips (see FAQ).
At first look, the listing of cons appears longer than the listing of professionals, however don’t be discouraged — as proven afterward, the enhancements in mannequin latency may be vital and price it.
Choice #1: Set up from supply
As talked about above, the ONNX runtime requires compatibility between many items of software program. If you wish to be on the innovative, one of the best ways to get the most recent model is to comply with the directions within the official Github repository. For Steady Diffusion particularly, this folder incorporates set up directions and pattern scripts for producing pictures. Anticipate constructing from supply to take fairly some time (round half-hour).
On the time of writing (Might 2024), this answer labored seamlessly for me on an Amazon EC2 occasion (g5.2xlarge, which comes with a A10G GPU). It avoids compatibility points mentioned beneath through the use of a Docker picture that comes with the best dependencies.
Choice #2: Set up by way of PyPI
In manufacturing, you’ll more than likely need a steady model of the ONNX runtime from PyPI, as an alternative of putting in the most recent model from supply. For Python particularly, there are two totally different libraries (one for CPU and one for GPU). Right here is the command to put in it for CPU:
pip set up onnxruntime
And right here is the command to put in it for GPU:
pip set up onnxruntime-gpu
You need to by no means set up each. Having them each may result in error messages or behaviors that aren’t straightforward to trace again to this root trigger. The ONNX runtime may merely fail to acknowledge the presence of the GPU, which is able to look stunning provided that onnxruntime-gpu is certainly put in.
In an excellent world, pip set up onnxruntime-gpu can be the tip of the story. Nevertheless, in follow, there are sturdy compatibility necessities between different items of software program in your machine, together with the working system, the hardware-specific drivers, and the Python model.
Say that you simply need to use the most recent model of the ONNX runtime (1.17.1) on the time of writing. So what stars do we have to align to make this occur?
Listed here are a few of the most typical sources of incompatibility that may enable you to arrange your atmosphere. The precise particulars will shortly turn into out of date, however the high-level concepts ought to proceed to use for some time.
CUDA compatibility
If you’re not planning on utilizing an NVIDIA GPU, you possibly can skip this part. CUDA is a platform for parallel computing that sits on high of NVIDIA GPUs, and is required for machine studying workflows. Every model of the ONNX runtime is suitable with solely sure CUDA variations, as you possibly can see in this compatibility matrix.
In accordance with this matrix, the most recent ONNX runtime model (1.17) is suitable with each CUDA 11.8 and CUDA 12. However you want to take note of the superb print: by default, ONNX runtime 1.17 expects CUDA 11.8. Nevertheless, most VMs as we speak (Might 2024) include CUDA 12.1 (you possibly can test the model by operating nvcc --version). For this specific setup, you’ll have to exchange the standard pip set up onnxruntime-gpu with:
pip set up onnxruntime-gpu==1.17.1 --extra-index-url https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/onnxruntime-cuda-12/pypi/easy/
Word that, as an alternative of being on the mercy of no matter CUDA model occurs to be put in in your machine, a cleaner answer is to do your work from inside a Docker container. You merely select the picture that has your required model of Python and CUDA. As an example:
docker run --rm -it --gpus all nvcr.io/nvidia/pytorch:23.10-py3
OS + Python + pip compatibility
This part discusses compatibility points which might be architecture-agnostic (i.e. you’ll encounter them whatever the goal accelerator). It boils down to creating certain that your software program (working system, Python set up and pip set up) are suitable together with your desired model of the ONNX runtime library.
Pip model: Until you’re working with legacy code or techniques, your most secure guess is to improve pip to the most recent model:
python -m pip set up --upgrade pip
Python model: As of Might 2024, the Python model that’s least possible to offer you complications is 3.10 (that is what most VMs include by default). Once more, except you’re working with legacy code, you definitely need at the very least 3.8 (since 3.7 was deprecated in June 2023).
Working system: The truth that the OS model also can hinder your capacity to put in the specified library got here as a shock to me, particularly that I used to be utilizing probably the most normal EC2 situations. And it wasn’t simple to determine that the OS model was the offender.
Right here I’ll stroll you thru my debugging course of, within the hopes that the workflow itself is longer-lived than the specifics of the variations as we speak. First, I put in onnxruntime-gpu with the next command (since I had CUDA 12.1 put in on my machine):
pip set up onnxruntime-gpu --extra-index-url https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/onnxruntime-cuda-12/pypi/easy/
On the floor, this could set up the most recent model of the library out there on PyPI. In actuality nevertheless, this can set up the most recent model suitable together with your present setup (OS + Python model + pip model). For me on the time, that occurred to be onnxruntime-gpu==1.16.0. (versus 1.17.1, which is the most recent). Unknowingly putting in an older model merely manifested within the ONNX runtime being unable to detect the GPU, with no different clues. After considerably by accident discovering the model is older than anticipated, I explicitly requested for the newer one:
pip set up onnxruntime-gpu==1.17.1 --extra-index-url https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/onnxruntime-cuda-12/pypi/easy/
This resulted in a message from pip complaining that the model I requested shouldn’t be really out there (regardless of being listed on PyPI):
ERROR: Couldn't discover a model that satisfies the requirement onnxruntime-gpu==1.17.1 (from variations: 1.12.0, 1.12.1, 1.13.1, 1.14.0, 1.14.1, 1.15.0, 1.15.1, 1.16.0, 1.16.1, 1.16.2, 1.16.3)
ERROR: No matching distribution discovered for onnxruntime-gpu==1.17.1
To grasp why the most recent model shouldn’t be getting put in, you possibly can go a flag that makes pip verbose: pip set up ... -vvv. This reveals all of the Python wheels that pip cycles by with the intention to discover the most recent one that’s suitable to your system. Here’s what the output seemed like for me:
Skipping hyperlink: not one of the wheel's tags (cp35-cp35m-manylinux1_x86_64) are suitable (run pip debug --verbose to indicate suitable tags): https://recordsdata.pythonhosted.org/packages/26/1a/163521e075d2e0c3effab02ba11caba362c06360913d7c989dcf9506edb9/onnxruntime_gpu-0.1.2-cp35-cp35m-manylinux1_x86_64.whl (from https://pypi.org/easy/onnxruntime-gpu/)
Skipping hyperlink: not one of the wheel's tags (cp36-cp36m-manylinux1_x86_64) are suitable (run pip debug --verbose to indicate suitable tags): https://recordsdata.pythonhosted.org/packages/52/f2/30aaa83bc9e90e8a919c8e44e1010796eb30f3f6b42a7141ffc89aba9a8e/onnxruntime_gpu-0.1.2-cp36-cp36m-manylinux1_x86_64.whl (from https://pypi.org/easy/onnxruntime-gpu/)
Skipping hyperlink: not one of the wheel's tags (cp37-cp37m-manylinux1_x86_64) are suitable (run pip debug --verbose to indicate suitable tags): https://recordsdata.pythonhosted.org/packages/a2/05/af0481897255798ee57a242d3989427015a11a84f2eae92934627be78cb5/onnxruntime_gpu-0.1.2-cp37-cp37m-manylinux1_x86_64.whl (from https://pypi.org/easy/onnxruntime-gpu/)
Skipping hyperlink: not one of the wheel's tags (cp35-cp35m-manylinux1_x86_64) are suitable (run pip debug --verbose to indicate suitable tags): https://recordsdata.pythonhosted.org/packages/17/cb/0def5a44db45c6d38d95387f20057905ce2dd4fad35c0d43ee4b1cebbb19/onnxruntime_gpu-0.1.3-cp35-cp35m-manylinux1_x86_64.whl (from https://pypi.org/easy/onnxruntime-gpu/)
Skipping hyperlink: not one of the wheel's tags (cp36-cp36m-manylinux1_x86_64) are suitable (run pip debug --verbose to indicate suitable tags): https://recordsdata.pythonhosted.org/packages/a6/53/0e733ebd72d7dbc84e49eeece15af13ab38feb41167fb6c3e90c92f09cbb/onnxruntime_gpu-0.1.3-cp36-cp36m-manylinux1_x86_64.whl (from https://pypi.org/easy/onnxruntime-gpu/)
...
The tags listed in brackets are Python platform compatibility tags, and you may learn extra about them right here. In a nutshell, each Python wheel comes with a tag that signifies what system it may well run on. As an example, cp35-cp35m-manylinux1_x86_64 requires CPython 3.5, a set of (older) Linux distributions that fall below the manylinux1 umbrella, and a 64-bit x86-compatible processor.
Since I needed to run Python 3.10 on a Linux machine (therefore filtering for cp310.*manylinux.*, I used to be left with a single potential wheel for the onnxruntime-gpu library, with the next tag:
cp310-cp310-manylinux_2_28_x86_64
You will get a listing of tags which might be suitable together with your system by operating pip debug --verbose. Here’s what a part of my output seemed like:
cp310-cp310-manylinux_2_26_x86_64
cp310-cp310-manylinux_2_25_x86_64
cp310-cp310-manylinux_2_24_x86_64
cp310-cp310-manylinux_2_23_x86_64
cp310-cp310-manylinux_2_22_x86_64
cp310-cp310-manylinux_2_21_x86_64
cp310-cp310-manylinux_2_20_x86_64
cp310-cp310-manylinux_2_19_x86_64
cp310-cp310-manylinux_2_18_x86_64
cp310-cp310-manylinux_2_17_x86_64
...
In different phrases, my working system is only a tad too outdated (the utmost linux tag that it helps is manylinux_2_26, whereas the onnxruntime-gpu library’s solely Python 3.10 wheel requires manylinux_2_28. Upgrading from Ubuntu 20.04 to Ubuntu 24.04 solved the issue.
As soon as the ONNX runtime is (lastly) put in, producing pictures with Steady Diffusion requires two following steps:
- Export the PyTorch mannequin to ONNX (this will take > half-hour!)
- Cross the ONNX mannequin and the inputs (textual content immediate and different parameters) to the ONNX runtime.
Choice #1: Utilizing official scripts from Microsoft
As talked about earlier than, utilizing the official pattern scripts from the ONNX runtime repository labored out of the field for me. In case you comply with their set up directions, you gained’t even should take care of the compatibility points talked about above. After set up, producing a picture is an easy as:
python3 demo_txt2img_xl.py "starry evening over Golden Gate Bridge by van gogh"
Beneath the hood, this script defines an SDXL mannequin utilizing Hugging Face’s diffusers library, exports it to ONNX format (which may take as much as half-hour!), then invokes the ONNX runtime.
Choice #2: Utilizing Hugging Face’s Optimum library
The Optimum library guarantees a whole lot of comfort, permitting you to run fashions on numerous accelerators whereas utilizing the acquainted pipeline APIs from the well-known transformers and diffusers libraries. For ONNX particularly, that is what inference code for SDXL appears like (extra in this tutorial):
from optimum.onnxruntime import ORTStableDiffusionXLPipelinemodel_id = "stabilityai/stable-diffusion-xl-base-1.0"
base = ORTStableDiffusionXLPipeline.from_pretrained(model_id)
immediate = "crusing ship in storm by Leonardo da Vinci"
picture = base(immediate).pictures[0]
# Remember to avoid wasting the ONNX mannequin
save_directory = "sd_xl_base"
base.save_pretrained(save_directory)
In follow, nevertheless, I struggled lots with the Optimum library. First, set up is non-trivial; naively following the set up instruction within the README file will run into the incompatibility points defined above. This isn’t Optimum’s fault per se, nevertheless it does add one more layer of abstraction on high of an already brittle setup. The Optimum set up may pull a model of onnxruntime that’s conflicting together with your setup.
Even after I gained the battle towards compatibility points, I wasn’t capable of run SDXL inference on GPU utilizing Optimum’s ONNX interface. The code snippet above (immediately taken from a Hugging Face tutorial) fails with some form mismatches, maybe resulting from bugs within the PyTorch → ONNX conversion:
[ONNXRuntimeError] : 1 : FAIL : Non-zero standing code returned whereas operating Add node.
Title:'/down_blocks.1/attentions.0/Add'
Standing Message: /down_blocks.1/attentions.0/Add: left operand can't broadcast on dim 3 LeftShape: {2,64,4096,10}, RightShape: {2,640,64,64}
For a quick second I thought of moving into the weeds and debugging the Hugging Face code (at the very least it’s open supply!), however gave up once I realized that Optimum has a backlog of greater than 250 points, with points going for weeks with no acknowledgement from the Hugging Face crew. I made a decision to maneuver on and easily use Microsoft’s official scripts as an alternative.
As promised, the hassle to get the ONNX runtime working is value it. On an A100 GPU, the inference time is decreased from 7–8 seconds (when operating vanilla PyTorch) to ~2 seconds. That is corresponding to TensorRT (an NVIDIA-specific different to ONNX), and about 1 second quicker than torch.compile (PyTorch’s native JIT compilation).
Reportedly, switching to much more performant GPUs (e.g. H100) can result in even increased beneficial properties from operating your mannequin with a specialised runtime.
The ONNX runtime guarantees vital latency beneficial properties, nevertheless it comes with non-trivial engineering overhead. It additionally faces the traditional trade-off for static compilation: inference is lots quicker, however the graph can’t be dynamically modified (which is at odds with dynamic adapters like peft). The ONNX runtime and related compilation strategies are value including to your pipeline when you’ve handed the experimentation part, and are able to put money into environment friendly manufacturing code.
In case you’re curious about optimizing inference time, listed below are some articles that I discovered useful:
[ad_2]