Starlings.Runtime.OnnxWeb.WebGpu
0.5.0
dotnet add package Starlings.Runtime.OnnxWeb.WebGpu --version 0.5.0
NuGet\Install-Package Starlings.Runtime.OnnxWeb.WebGpu -Version 0.5.0
<PackageReference Include="Starlings.Runtime.OnnxWeb.WebGpu" Version="0.5.0" />
<PackageVersion Include="Starlings.Runtime.OnnxWeb.WebGpu" Version="0.5.0" />
<PackageReference Include="Starlings.Runtime.OnnxWeb.WebGpu" />
paket add Starlings.Runtime.OnnxWeb.WebGpu --version 0.5.0
#r "nuget: Starlings.Runtime.OnnxWeb.WebGpu, 0.5.0"
#:package Starlings.Runtime.OnnxWeb.WebGpu@0.5.0
#addin nuget:?package=Starlings.Runtime.OnnxWeb.WebGpu&version=0.5.0
#tool nuget:?package=Starlings.Runtime.OnnxWeb.WebGpu&version=0.5.0
Starlings.Runtime.OnnxWeb.WebGpu
The voluntary WebGPU adapter of Starlings.Runtime.OnnxWeb. It ships onnxruntime-web 1.30.0's WebGPU-capable bundle (ort.webgpu.min.mjs and its asyncify binary, 26.8 MB, about 4 MB brotli) as static web assets, and OnnxWebGpuBundle.WebGpu to initialize with. No core package references it; reference it only if you want WebGPU.
Maturity
Maturity is per profile, generated from the registry manifests by model-pipeline catalog (do not edit this table by hand). Gates, evidence and capabilities per profile: the model catalog.
| Profile | Maturity | Platforms |
|---|---|---|
multilingual-e5-small/onnx-fp32-webgpu |
experimental | linux-x64 webgpu chromium-153: experimental; linux-x64 webgpu firefox: untested; macos-arm64 webgpu safari: untested; windows-x64 webgpu chromium: untested |
multilingual-e5-small/onnx-int8embed-webgpu |
experimental | linux-x64 webgpu chromium-153: experimental; linux-x64 webgpu firefox: untested; macos-arm64 webgpu safari: untested; windows-x64 webgpu chromium: untested |
Experimental. WebGPU stays local evidence (no CI job). Through this package, both e5 WebGPU profiles pass every bound of their protected tolerances, G1 and G2 included, on one NVIDIA RTX 4060 Ti in headless Chromium 153 (Chrome for Testing) on Linux: fp32 within 2.05e-7 of the reference, int8-embed 3.51e-3 with minimum cosine 0.99980 and 1.7e-7 from native ONNX Runtime on the same graph (evidence). fp16 graphs need the adapter feature shader-f16, which that setup does not expose. Other GPUs, vendors, browsers and operating systems are untested.
Use
using Starlings.Runtime.OnnxWeb;
using Starlings.Runtime.OnnxWeb.WebGpu;
await OnnxWebRuntime.InitializeAsync(new OnnxWebRuntimeOptions { Bundle = OnnxWebGpuBundle.WebGpu });
await using var session = await OnnxWebSession.CreateAsync(model, new OnnxWebSessionOptions { Provider = OnnxWebExecutionProvider.WebGpu });
A page loads one bundle. This one also runs the WASM provider, so an app that offers both providers initializes with it. When navigator.gpu is missing, CreateAsync throws NotSupportedException and never falls back to WASM on its own. The first run compiles shaders (0.3 to 0.7 s); after that, a call costs about 17 ms whatever the batch size, which wins on batches and long texts.
| Product | Versions Compatible and additional computed target framework versions. |
|---|---|
| .NET | net10.0 is compatible. net10.0-android was computed. net10.0-browser was computed. net10.0-ios was computed. net10.0-maccatalyst was computed. net10.0-macos was computed. net10.0-tvos was computed. net10.0-windows was computed. |
-
net10.0
- Starlings.Runtime.OnnxWeb (= 0.5.0)
NuGet packages
This package is not used by any NuGet packages.
GitHub repositories
This package is not used by any popular GitHub repositories.