Add onnxruntim to support inference of various models.

[yst-one]

This library is widely used, especially in this booming AI era.
https://github.com/microsoft/onnxruntime

[ndonkoHenri]

Thanks for the request! I looked into whether we could ship an onnxruntime wheel and unfortunately it's not something we can reasonably build right now. A few reasons:

• No source distribution: onnxruntime ships only prebuilt binary wheels for desktop/server platforms (macOS, manylinux, Windows) — there's no sdist on PyPI, and no iOS/Android wheels. So a mobile build would mean cross-compiling the entire ONNX Runtime C++ tree from GitHub source.
• Upstream's mobile story is native-only, by design: Microsoft ships mobile as native SDKs — the onnxruntime-android AAR (Maven) and the onnxruntime-c / onnxruntime-objc CocoaPods — meant to be consumed from Java/Kotlin/Swift/C++.

We support pyjnius/pyobjus - maybe it could of help when calling native ORT SDK from your code. But onnxruntime itself cant be supported atm.

[yst-one]

First of all, thank you for your reply. Oh I see. If there's no source code available on PyPI, compilation won't be possible. Have you considered ncnn? It's a library optimized specifically for mobile devices for running inference models. https://pypi.org/project/ncnn/#files

[yst-one]

This is a library for running models on mobile platforms maintained by Tencent, a Chinese company. It enables models to run more efficiently on mobile and embedded devices. It also boasts an active community, and nearly all mainstream models can be converted to work with ncnn.

[yst-one]

I previously forked your Mobile Forge project and tried to compile it, but I never managed to get it to build successfully. You’re really skilled and knowledgeable about compiling these packages. I’d love to learn from you, and it would be amazing if you could release a documentation on how to edit recipes on flet.dev, haha.