Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/happyme531/ztu_somemodelruntime_ez_rknn_async/llms.txt

Use this file to discover all available pages before exploring further.

EZ RKNN Async is a native extension that must be compiled on (or for) the target Rockchip device. There is no pre-built wheel on PyPI — you build a wheel from source and install it locally.

​
Requirements

Before building, make sure the following are present on your device:
  • Python 3.7+ with pip
  • Rockchip device with RKNPU2 (RK3588, RK3566, RK3568, etc.)
  • librknnrt.so installed system-wide (provided by the Rockchip RKNN SDK)
  • CMake 3.18+
  • A C++17-capable compiler (GCC 8+ or Clang 7+)
  • pybind11 ≥ 2.10 (installed automatically by the build system)
RKNN SDK version 2.4.1 or later is strongly recommended. If an older version of librknnrt.so is detected at runtime, the library will emit a warning that behavior may be unstable.

​
Build and install from source

1

Clone the repository

git clone https://github.com/happyme531/ztu_somemodelruntime_ez_rknn_async.git
cd ztu_somemodelruntime_ez_rknn_async
2

Build a wheel

Use pip’s wheel-building mode with scikit-build-core as the build backend. The --no-deps flag skips building dependency wheels, keeping the output directory clean.
python3 -m pip wheel . -w dist --no-deps
CMake 3.18+ is required. The build system picks up the RKNN headers and librknnrt.so from the system include and library paths automatically.
To build with debug symbols for development, add --config-settings=cmake.build-type=Debug to the command above.
3

Install the built wheel

pip install dist/*.whl
This installs the compiled extension along with its Python runtime dependencies (numpy and, on Python 3.7, typing_extensions).
4

Verify the installation

import ztu_somemodelruntime_ez_rknn_async as ez
print(ez.__version__)
print(ez.InferenceSession)
If the import succeeds and prints the version string, the library is installed correctly.

​
Build dependencies summary

DependencyMinimum versionNotes
Python3.7
CMake3.18Required by scikit-build-core
C++ compilerC++17GCC 8+ or Clang 7+
pybind112.10Fetched automatically if not present
scikit-build-core0.5 (Python < 3.8), 0.7 (Python ≥ 3.8)Build backend
numpy1.21Runtime dependency
typing_extensions4.0Runtime dependency on Python 3.7 only
librknnrt.so2.4.1+ recommendedMust be installed on the device

​
Troubleshooting

librknnrt.so not found during build Make sure the RKNN SDK runtime library is installed. On most Linux distributions this means the .so file is in /usr/lib or /usr/local/lib. If it is in a non-standard location, set LD_LIBRARY_PATH before building: 
export LD_LIBRARY_PATH=/path/to/rknn/lib:$LD_LIBRARY_PATH
CMake version too old The default system CMake on older distributions may be below 3.18. Install a newer version via pip: 
pip install cmake
C++17 errors Pass the standard explicitly if your compiler does not default to C++17: 
python3 -m pip wheel . -w dist --no-deps \
  --config-settings=cmake.args="-DCMAKE_CXX_STANDARD=17"

Build docs developers (and LLMs) love

Get started for freeTalk to us