Describe how the official libraries were built

[ethindp]

As it currently stands, I cannot find any way of reproducing the binary files that are in the repository. The code requires that the actual ONNX model be at a hardcoded path, but the Python wheels and compiled shared libraries seem to not have this requirement at all, from what I can tell. Was the source code changed to universally hard-code this path sometime after the libraries were built?

[smilexin]

Clarification on the Binary Files and ONNX Model Path

Thanks for the thoughtful analysis—there is a common point of confusion here, so let me clarify.

This repository actually includes two different open-source options, in different directories:

• Precompiled library option: the ONNX model and inference logic are already embedded inside the shared library, so no external ONNX path is required at runtime. Can refer to include/lib/examples.
• Source-code option: you need to use the ONNX model yourself and compile it together with the pre/post‑processing code using a third‑party inference engine (e.g., ORT). Can refer to src/examples_onnx.

It’s simply because the two options have different scopes. You can follow the source build demo in examples_onnx we provide to reproduce the full source‑based workflow.

Good luck!