LMRuntime.com / Current state

Project Status

A plain-language view of the expanded public package family, generated documentation, backend execution status, version policy, and evidence boundaries that remain application-specific.

The expanded UAIX.LmRuntime package family is documented as public NuGet artifacts with explicit separation between available managed execution, diagnostic backend declarations, RID-specific package slots, and claims that require separate proof.

Works Today

Public package family17 package guides

Every package guide has version-free install forms, exact dependencies, deep examples, NuGet links, and a current ownership boundary.

Application entry pointUAIX.LmRuntime.LocalEndpoint

Verified local GGUF loading, isolated sessions, managed CPU execution, bounded generation, and explicit capability evidence.

Backend selection surfaceAcceleration + CpuManaged

Explicit registration, local probing, selection policies, device identity, diagnostics, and recorded CPU fallback.

Generated reference376 types / 2,044 members

Managed public APIs are represented from XML documentation; RID-only packages are listed with zero managed public types.

Backend execution status

PackageCurrent supplied behaviorGuide / NuGet
Backends.CpuManagedRegisters a managed CPU backend and probes available with an explicit CPU device descriptor.Guide · NuGet ↗
Backends.CudaRegisters CUDA identity and compatibility diagnostics; probes unavailable until native asset, runtime, driver, and device proof exists.Guide · NuGet ↗
Backends.DirectMLRegisters the Windows DirectML compatibility lane; supplied diagnostic implementation probes unavailable.Guide · NuGet ↗
Backends.VulkanRegisters a vendor-diverse Vulkan compatibility lane; supplied diagnostic implementation probes unavailable.Guide · NuGet ↗
Backends.RocmRegisters the AMD ROCm compatibility lane; supplied diagnostic implementation probes unavailable.Guide · NuGet ↗
Backends.MetalRegisters the Apple Metal compatibility lane; supplied diagnostic implementation probes unavailable.Guide · NuGet ↗
CUDA RID packagesReserve Windows x64 package-visible native asset paths and currently contain marker files, not CUDA inference binaries.Modern · K80

Documentation status

All package pages include purpose, audience, execution state where applicable, install commands, dependencies with guide and NuGet links, workflow, package boundaries, source-derived examples, FAQs, and API or RID-asset documentation. Machine-readable catalog and API JSON files mirror the public pages.

Version policy

LMRuntime.com does not embed UAIX.LmRuntime package version numbers in install commands, comparisons, or API headings. NuGet owns current versions, target frameworks, hashes, release metadata, and resolved dependency versions. Applications should pin dependencies through their normal central package and lock-file process.

Not Claimed

Active GPU inference from diagnostic packagesRegistration and capability metadata are not proof that CUDA, DirectML, Vulkan, ROCm, or Metal executed model math.
Embedded CUDA binaries in RID slotsThe supplied Windows native packages contain marker files rather than inference DLLs.
Broad model compatibilityGGUF validity, tokenizer behavior, model architecture, tensor storage, and kernel coverage remain separate checks.
Benchmark leadershipNo speed, throughput, latency, memory, or energy claim is made without reproducible retained evidence.
Model downloads or provider accessThe package family does not imply a hosted service, model catalog, provider client, or network fallback.
Production or safety certificationEach application owns deployment review, security controls, supported-model declarations, operations, and support terms.