Build onnxruntime wheel with QNN backend for aarch64 (Linux)

This builds onnxruntime wheels for aarch64 on Linux, with bindings to Qualcomm AI Engine Direct to run ONNX models on NPU/GPU on Dragonwing hardware.

To build this wheel you'll need:

1. An aarch64/arm64 host machine, for example on macOS (Apple silicon).

2. Docker.

3. Open a command prompt or terminal window and run:

   sh build.sh

You'll find the output in out/.

Changing Python or onnxruntime version

See the comments in Dockerfile

Prebuilt wheels

• onnxruntime_qnn-1.22.2-cp312-cp312-linux_aarch64.whl
• onnxruntime_qnn-1.23.0-cp312-cp312-linux_aarch64.whl