aikernel-cuda13-libtorch2-12-win-x64

日本語

Python wrapper package for the AIKernel CUDA 13.0 + LibTorch 2.12 Windows win-x64 Capability.

Install the stable package from PyPI:

pip install aikernel-cuda13-libtorch2-12-win-x64

Import it as:

import aikernel_cuda13_libtorch2_12_win_x64 as cuda_capability

descriptor = cuda_capability.capability_descriptor()
print(descriptor["package_id"])

This Python package is intentionally lightweight and is distributed through pip. Release wheels include the CUDA Capability managed assembly, libtorch_bridge.dll, and loader.json. They do not include LibTorch, CUDA runtime DLLs, cuDNN, or cuBLAS. The executable CUDA runtime snapshot is distributed as the full GitHub Release runtime archive.

Use this package to discover the Capability identity, supported runtime, and installation guidance from Python tooling. NuGet is reserved for C# consumers; Python wrappers are not embedded in NuGet packages.

The v0.1.2 package also exposes the generated managed API catalog through managed_api_catalog(), managed_api_summary(), managed_type_names(), and find_managed_type(full_name).

The package includes a bundled loader.json template and loader helpers:

config = cuda_capability.load_loader_config()
print(config.native_library)
print(config.resolved_runtime_search_paths())

This is lightweight runtime support only. The pip package ships the AIKernel managed Capability assembly and native bridge, but does not ship LibTorch, CUDA, cuDNN, or cuBLAS runtime binaries.

Release wheels expose bundled paths when those binaries are present:

print(cuda_capability.bundled_managed_assemblies())
print(cuda_capability.bundled_native_libraries())

Development wheels are published as GitHub Release assets, not to PyPI. PyPI is reserved for stable releases.