wonnx
DirectML
Our great sponsors
wonnx | DirectML | |
---|---|---|
18 | 26 | |
1,487 | 1,936 | |
6.8% | 4.5% | |
6.5 | 7.4 | |
26 days ago | 10 days ago | |
Rust | Python | |
GNU General Public License v3.0 or later | 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.
wonnx
-
Intel CEO: 'The entire industry is motivated to eliminate the CUDA market'
The two I know of are IREE and Kompute[1]. I'm not sure how much momentum the latter has, I don't see it referenced much. There's also a growing body of work that uses Vulkan indirectly through WebGPU. This is currently lagging in performance due to lack of subgroups and cooperative matrix mult, but I see that gap closing. There I think wonnx[2] has the most momentum, but I am aware of other efforts.
[1]: https://kompute.cc/
[2]: https://github.com/webonnx/wonnx
-
VkFFT: Vulkan/CUDA/Hip/OpenCL/Level Zero/Metal Fast Fourier Transform Library
To a first approximation, Kompute[1] is that. It doesn't seem to be catching on, I'm seeing more buzz around WebGPU solutions, including wonnx[2] and more hand-rolled approaches, and IREE[3], the latter of which has a Vulkan back-end.
[1]: https://kompute.cc/
[2]: https://github.com/webonnx/wonnx
[3]: https://github.com/openxla/iree
-
Onnx Runtime: “Cross-Platform Accelerated Machine Learning”
There's also a third-party WebGPU implementation: https://github.com/webonnx/wonnx
-
Are there any ML crates that would compile to WASM?
By experimental I meant e.g. using WGPU to run compute shaders like wonnx, which is working fine but only on a very restricted set of devices and browsers.
- WebGPU ONNX inference runtime written in Rust
-
PyTorch Primitives in WebGPU for the Browser
https://news.ycombinator.com/item?id=35696031 ... TIL about wonnx: https://github.com/webonnx/wonnx#in-the-browser-using-webgpu...
microsoft/onnxruntime: https://github.com/microsoft/onnxruntime
Apache/arrow has language-portable Tensors for cpp: https://arrow.apache.org/docs/cpp/api/tensor.html and rust: https://docs.rs/arrow/latest/arrow/tensor/struct.Tensor.html and Python: https://arrow.apache.org/docs/python/api/tables.html#tensors https://arrow.apache.org/docs/python/generated/pyarrow.Tenso...
Fwiw it looks like the llama.cpp Tensor is from ggml, for which there are CUDA and OpenCL implementations (but not yet ROCm, or a WebGPU shim for use with emscripten transpilation to WASM): https://github.com/ggerganov/llama.cpp/blob/master/ggml.h
Are the recommendable ways to cast e.g. arrow Tensors to pytorch/tensorflow?
FWIU, Rust has a better compilation to WASM; and that's probably faster than already-compiled-to-JS/ES TensorFlow + WebGPU.
What's a fair benchmark?
-
rustformers/llm: Run inference for Large Language Models on CPU, with Rust 🦀🚀🦙
wonnx has done some fantastic work in this regard, so that's where we plan to start once we get there. In terms of general discussion of alternate backends, see this issue.
-
I want to talk about WebGPU
> GPU in other ways, such as training ML models and then using them via an inference engine all powered by your local GPU?
Have a look at wonnix https://github.com/webonnx/wonnx
A WebGPU-accelerated ONNX inference run-time written 100% in Rust, ready for native and the web
-
Chrome Ships WebGPU
Looking forward to your WebGPU ML runtime! Also, why not contribute back to WONNX? (https://github.com/webonnx/wonnx)
-
OpenXLA Is Available Now
You can indeed perform inference using WebGPU (see e.g. [1] for GPU-accelerated inference of ONNX models on WebGPU; I am one of the authors).
The point made above is that WebGPU can only be used for GPU's and not really for other types of 'neural accelerators' (like e.g. the ANE on Apple devices).
[1] https://github.com/webonnx/wonnx
DirectML
- Microsoft DirectML: high-performance DirectX 12 library for ML
-
AMD Radeon RX 7600 XT Linux Performance
Only reason I am using the DirectML fork of Automatic1111 is because I am on Windows and pytorch hasn't caught up to RocM 6.
DirectML is fully supported path on Windows and is support by Microsoft et al. (https://github.com/microsoft/DirectML).
Everyone is moving off Cuda as quickly as possible not because the other are better, per se, but because it is easier and cheaper.
-
Train issue on AMD card
See: https://github.com/microsoft/DirectML/issues/400
-
'Everyone and Their Dog is Buying GPUs,' Musk Says as AI Startup Details Emerge
ONNX (https://onnx.ai/ https://github.com/onnx/onnx) is an alternative to the basic CUDA model, using Direct-ML ( https://learn.microsoft.com/en-us/windows/ai/directml/dml-intro https://github.com/microsoft/DirectML), which is a microsoft-backed open approach. That is what has allowed AMD cards, even slightly older ones, to join in on the machine learning fun.
-
AMD ROCm: A Wasted Opportunity
It's really shocking that AMD fails to extend support natively.
Workarounds such as DirectML claim to be the answer in unifying people with NVIDIA or AMD GPUs, but thus far it hasn't, with issues such as [this](https://github.com/microsoft/DirectML/issues/58) constantly popping up.
As nicolaslem points out, Arch does have community packages for ROCm, but that, unsurprisingly fails to lend support to many consumer GPUs. The best community support I have come across are [rocm-opencl](https://copr.fedorainfracloud.org/coprs/mystro256/rocm-openc... [rocm-hip](https://copr.fedorainfracloud.org/coprs/mystro256/rocm-hip/) for Fedora maintained by [mystro256](https://github.com/Mystro256), who is a single AMD employee.Thanks to him, my AMD GPU (Radeon 6800XT) hasn't completely gone to waste, and I was able to tinker with some things (Gaming isn't really up my alley).
Lately however, after beginning to work on DGX V100s and A100s, and using my older laptop with a GTX 1650, it was apparent how simple setting up CUDA was, and how easily I could experiment with it on my consumer card. Many have spoken about similar stories, and here's mine. Really hope AMD does a whole lot more, and doesn't exclusively keep their powerful GPUs for gaming.
-
nn-morse neural network mentioned in ftroop by VK6MIK
You can use a NVidia gpu or the cpu to do the training but the cpu training is very very slow. For AMD graphics cards like the AMD Radeon VII the only solution is pytorch_directml but unfortunately there appears to be a bug that stops it working nn-morse and torch-directml memory leak? · Issue #355
- Trying to get my computer set up for ML
- ROCm installation on Acer Aspire 3
-
Microsoft’s PyTorch-DirectML Release-2 Now Works with Python Versions 3.6, 3.7, 3.8, and Includes Support for GPU Device Selection to Train Machine Learning Models
Github: https://github.com/microsoft/DirectML
- Dying Light 2 is 30 fps on Series S 😴
What are some alternatives?
stablehlo - Backward compatible ML compute opset inspired by HLO/MHLO
onnx - Open standard for machine learning interoperability
text2image-gui - Somewhat modular text2image GUI, initially just for Stable Diffusion
tract - Tiny, no-nonsense, self-contained, Tensorflow and ONNX inference
civitai - A repository of models, textual inversions, and more
iree - A retargetable MLIR-based machine learning compiler and runtime toolkit.
burn - Burn is a new comprehensive dynamic Deep Learning Framework built using Rust with extreme flexibility, compute efficiency and portability as its primary goals.
blaze - A Rustified OpenCL Experience
benchmark - TorchBench is a collection of open source benchmarks used to evaluate PyTorch performance.
burn - Burn is a new comprehensive dynamic Deep Learning Framework built using Rust with extreme flexibility, compute efficiency and portability as its primary goals. [Moved to: https://github.com/Tracel-AI/burn]
gpuweb - Where the GPU for the Web work happens!