UAIX.LmRuntime / Package guide

UAIX.LmRuntime.Backends.Cuda

CUDA backend registration, declared compatibility metadata, runtime identifiers, and fail-closed local probe diagnostics.

Required For CUDA backend registration and fail-closed CUDA diagnostics

UAIX.LmRuntime.Backends.Cuda

CUDA backend registration, declared compatibility metadata, runtime identifiers, and fail-closed local probe diagnostics.

Open on NuGet Package family

Overview

Backends.Cuda registers the UAIX CUDA backend family with Acceleration. It declares the intended CUDA capability surface and package runtime identifiers (win-x64, linux-x64), then probes unavailable until a host-supplied native adapter proves assets, runtime libraries, drivers, and a device. The managed package performs no hidden native inference. Modern Windows and Tesla K80 native-asset slots are published as separate package identities.

Who should use it Hosts building an explicit CUDA compatibility lane, diagnostics UI, deployment validation, or future native adapter integration for NVIDIA GPU hosts.

Install

.NET CLI

dotnet add package UAIX.LmRuntime.Backends.Cuda

Project file

<PackageReference Include="UAIX.LmRuntime.Backends.Cuda" />

Version policy: The documentation deliberately omits UAIX.LmRuntime package version numbers. Resolve and pin versions through your normal dependency-management and lock-file process.

Review the current package metadata, frameworks, dependencies, and downloads on NuGet ↗

Direct package dependencies

UAIX.LmRuntime.AccelerationNuGet ↗

Package role and boundaries

Required For CUDA backend registration and fail-closed CUDA diagnostics

You need the stable CUDA backend ID and registry extension.
You need package-visible CUDA runtime-identifier and capability declarations.
You need fail-closed probe diagnostics before CUDA execution is proven.

Boundary

Installing or registering this package does not prove CUDA execution.
The supplied managed backend does not load native inference libraries or enumerate a real device.
Do not report GPU availability unless a host adapter returns an available probe with concrete device evidence.

Registration is not execution

AddUaixCudaBackend adds metadata and probe behavior to the registry; it does not start a CUDA engine.

Compatibility fails closed

The default diagnostic probe returns unavailable and states which native evidence has not been supplied.

Runtime identifiers are declared

The package declares win-x64, linux-x64; the current process and deployment still require independent validation.

Key types

These are the main entry points for this package. The generated reference below includes every documented type and member represented by public package XML documentation.

CudaRuntimeBackend RuntimeBackendRegistryExtensions

Coding examples

Examples use public package signatures documented on LMRuntime.com. Model paths, hashes, byte counts, prompts, and host-specific identifiers remain application inputs.

Register the CUDA backend

Add the package-visible CUDA backend to an explicit Acceleration registry.

RegisterCUDAExample.cs

using UAIX.LmRuntime.Acceleration;
using UAIX.LmRuntime.Backends.Cuda;

var registry = new RuntimeBackendRegistry();
registry.AddUaixCudaBackend();

IRuntimeBackend backend = registry.FindById(CudaRuntimeBackend.BackendId)
    ?? throw new InvalidOperationException("CUDA backend registration failed.");

Console.WriteLine($"{backend.Id} / {backend.Kind}");

Boundary: Registration exposes CUDA diagnostics. It does not prove a native runtime or device.

Probe the CUDA package boundary

Inspect declared capabilities and the independent availability result.

ProbeCUDAExample.cs

using UAIX.LmRuntime.Acceleration;
using UAIX.LmRuntime.Backends.Cuda;

var backend = new CudaRuntimeBackend();
RuntimeBackendProbeResult probe = await backend.ProbeAsync(
    new RuntimeBackendOptions
    {
        RequestedRuntimeIdentifier =
            System.Runtime.InteropServices.RuntimeInformation.RuntimeIdentifier,
        RequireNativeAssets = true
    });

Console.WriteLine($"Declared API: {probe.Capabilities.BackendApiName}");
Console.WriteLine($"Native state: {probe.Capabilities.NativeAssetState}");
Console.WriteLine($"Available: {probe.IsAvailable}");
foreach (string diagnostic in probe.Diagnostics)
{
    Console.WriteLine(diagnostic);
}

Boundary: The default probe is expected to remain unavailable until a production adapter proves CUDA assets, runtime libraries, and a device.

Require the CUDA backend by ID

Reject CPU substitution when the workload contract specifically requires CUDA.

RequireCUDAExample.cs

using UAIX.LmRuntime.Acceleration;
using UAIX.LmRuntime.Backends.Cuda;

var registry = new RuntimeBackendRegistry();
registry.AddUaixCudaBackend();

var selector = new RuntimeBackendSelector(registry);
RuntimeSelectionResult result = await selector.SelectAsync(
    new RuntimeBackendOptions
    {
        Policy = RuntimeSelectionPolicy.RequireBackendId,
        PreferredBackendId = CudaRuntimeBackend.BackendId,
        AllowCpuFallback = false,
        RequireNativeAssets = true
    });

if (!result.Succeeded)
{
    Console.Error.WriteLine(result.FailureReason);
}

Boundary: RequireBackendId fails instead of relabeling another backend as CUDA.

List the CUDA package runtime identifiers

Use declared identifiers as deployment metadata, not as proof that the current machine is ready.

CUDARuntimeIdentifiersExample.cs

using UAIX.LmRuntime.Backends.Cuda;

var backend = new CudaRuntimeBackend();
foreach (string runtimeIdentifier in backend.Capabilities.RuntimeIdentifiers)
{
    Console.WriteLine(runtimeIdentifier);
}

Boundary: The declared identifiers are win-x64, linux-x64. Probe evidence remains required for a usable CUDA execution path.

Open this package on NuGet Compare the full package family →

Generated API reference

Expand a type to review its documented fields, properties, constructors, methods, parameter descriptions, and return descriptions. Browser Find also works across the closed detail elements.

CudaRuntimeBackendUAIX.LmRuntime.Backends.Cuda 2 members

Reports CUDA backend compatibility and diagnostics to the acceleration registry.

This package does not contain hidden CUDA inference binaries. Native assets and device proof must come from a declared native package or host adapter before CUDA execution can be selected.

Field BackendId

The stable backend identifier used by selection policies.

Method CudaRuntimeBackend

Initializes a new instance of the UAIX.LmRuntime.Backends.Cuda.CudaRuntimeBackend class.

RuntimeBackendRegistryExtensionsUAIX.LmRuntime.Backends.Cuda 1 member

Provides registration helpers for the CUDA backend.

Registration only exposes diagnostics and selection metadata. It does not authorize hidden native execution.

Method AddUaixCudaBackend(UAIX.LmRuntime.Acceleration.IRuntimeBackendRegistry)

Adds the UAIX CUDA backend to a registry.

registry: The registry to update.

Returns: The same registry instance for fluent configuration.

Frequently asked questions

Does this package execute through CUDA today?

The managed registration package reports unavailable until a host-supplied native adapter proves CUDA assets, runtime libraries, drivers, and a device.

Why does SupportsGpuExecution read true when IsAvailable is false?

Capabilities describe the backend family. IsAvailable describes the current proven host state.

Does registration contact the network?

No. Registration and the supplied diagnostic probe perform no downloads, provider calls, subprocess execution, or remote inference.