wonnx
wonnx copied to clipboard
webonnx
A WebGPU-accelerated ONNX inference run-time written 100% in Rust, ready for native and the web
I have a first draft of a working ConvTranspose implementation. Still want to add support for padding and maybe groups before we merge this. Also should probably handle some more...
Hello 👋! I need the slice operator (v11) to infer one of our models. I have seen in issue #153 that the slice operator is not yet implemented in `wonnx`,...
Network weights can be stored as `f16` floats, halving the size of the network, which is often very desirable. It would be nice if wonnx could support loading networks that...
This is a first attempt at addressing #169 as well as an earlier suggestion to move ONNX out of the IR. tl;dr, this PR exposes a 'builder API' that allows...
**Is your feature request related to a problem? Please describe.** I'm one of the maintainers of the [llm](https://github.com/rustformers/llm) project, and we're looking for a robust, cross-platform GPU inferencing solutions for...
The shaders created by wonnx apparently run into naga/wgpu bugs that cause a lot of validation errors on Vulkan: ``` [2023-04-18T00:42:44Z ERROR wgpu_hal::vulkan::instance] VALIDATION [UNASSIGNED-CoreValidation-Shader-InconsistentSpirv (0x6bbb14)] Validation Error: [ UNASSIGNED-CoreValidation-Shader-InconsistentSpirv...
Work in progress... just the parser for now. Some useful resources: * https://onnx.ai/onnx/operators/onnx__Einsum.html * https://github.com/termoshtt/einsum-derive/blob/main/einsum-codegen/src/subscripts.rs#L1 * https://stackoverflow.com/questions/26089893/understanding-numpys-einsum
This PR contains #139 , which was pulled out of the master branch. We can merge this when @CameronTofer agrees to relicense their contribution as "MIT OR Apache-2.0" (see #150).