Themis.Inference.CoreML
4.0.2-preview
dotnet add package Themis.Inference.CoreML --version 4.0.2-preview
NuGet\Install-Package Themis.Inference.CoreML -Version 4.0.2-preview
<PackageReference Include="Themis.Inference.CoreML" Version="4.0.2-preview" />
<PackageVersion Include="Themis.Inference.CoreML" Version="4.0.2-preview" />
<PackageReference Include="Themis.Inference.CoreML" />
paket add Themis.Inference.CoreML --version 4.0.2-preview
#r "nuget: Themis.Inference.CoreML, 4.0.2-preview"
#:package Themis.Inference.CoreML@4.0.2-preview
#addin nuget:?package=Themis.Inference.CoreML&version=4.0.2-preview&prerelease
#tool nuget:?package=Themis.Inference.CoreML&version=4.0.2-preview&prerelease
Themis.Inference.CoreML
Themis.Inference.CoreML is an optional prototype that selects ONNX Runtime's
Core ML execution provider for local typed-decision inference. It targets
net10.0; the initial deployment assumption is a macOS 12+ Apple Silicon
consumer published for osx-arm64. Neural Engine device selection requires
macOS 13+; ONNX Runtime may expose only the GPU on earlier systems. No Apple
framework is referenced by this managed package, and no Core ML native runtime
is bundled.
Register it alongside AddThemis:
services.AddThemis(options => options.Model = "apple-model");
services.AddThemisCoreML();
The default configuration requires CoreMLExecutionProvider and an
accelerator. When ONNX Runtime exposes a Neural Engine device, the selection
policy prefers it; otherwise it selects the GPU. Session options match
MLComputeUnits to the selected device (CPUAndNeuralEngine or CPUAndGPU).
This does not guarantee that any particular model node runs on the accelerator;
ONNX Runtime may partition unsupported nodes to CPU.
The host must load an ONNX Runtime native library built with Core ML support
(--use_coreml) and compatible with the managed ONNX Runtime version resolved
by Themis. The stock .NET package's Core ML availability is not assumed, and
the prototype does not replace or redistribute the host's native runtime.
When Core ML is explicitly required, initialization rejects non-macOS or
non-arm64 hosts and reports when the loaded runtime does not expose the Core ML
provider. A caller-provided InferenceRuntimeConfiguration is preserved.
Package restore, build, and package creation are validated in CI. The package
is published with a -preview NuGet version as a prototype. There is no
Apple-hosted consumer, signing/notarization, native-runtime asset, model-shape,
or hardware-profile validation yet, and preview publication does not establish
Apple platform or model support.
References:
| 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
- Microsoft.Extensions.DependencyInjection.Abstractions (>= 10.0.13)
- Themis.Core (>= 4.0.2)
- Themis.Inference (>= 4.0.2)
NuGet packages
This package is not used by any NuGet packages.
GitHub repositories
This package is not used by any popular GitHub repositories.
| Version | Downloads | Last Updated |
|---|---|---|
| 4.0.2-preview | 43 | 10/27/2026 |
| 4.0.1-preview | 39 | 10/27/2026 |
| 4.0.0-preview | 46 | 10/26/2026 |
| 2.0.3-preview | 49 | 10/25/2026 |