Metadata-Version: 2.4
Name: holoscan-inference-cu13
Version: 5.0.0
Summary: Holoscan Inference bindings for CUDA 13
Author: NVIDIA Corporation
License-Expression: Apache-2.0
License-File: LICENSE.txt
License-File: NOTICE.txt
Classifier: Operating System :: POSIX :: Linux
Classifier: Programming Language :: Python :: 3.12
Classifier: Programming Language :: Python :: 3.13
Classifier: Programming Language :: Python :: 3.14
Classifier: Environment :: GPU :: NVIDIA CUDA :: 13
Requires-Python: >=3.12
Description-Content-Type: text/markdown
Requires-Dist: holoscan-core-cu13==5.0.0

# Holoscan Inference for CUDA 13

This distribution owns the `holoscan_inference` provider package, its native
extension, HoloInfer, and the Inference operator library. It requires the exact
matching `holoscan-core-cu13` wheel; Core libraries are not duplicated. The
`holoscan.modules` entry point exposes the provider as the convenient
`holoscan.inference` public alias.

TensorRT 10 and CUDA 13 remain external system runtime prerequisites. This
wheel deliberately does not redistribute those vendor libraries or select a
TensorRT Python distribution on the user's behalf.

Build from a completed SDK build tree using regular CPython 3.12:

```bash
export CUDA_MAJOR=13 PEP440_VERSION=5.0.0.dev1
python3.12 -m build --wheel python/holoscan-inference \
  -Ccmake.define.HOLOSCAN_SDK_BUILD_ROOT=/absolute/path/to/sdk/build
```

The shared scikit-build-core project stages the completed native build without
reconfiguring or compiling the SDK. The wheel targets `cp312-abi3` and is
qualified on regular CPython 3.12–3.14. Uninstalling Inference leaves Core's
files intact.
