microsoft/onnxruntime ? reverse-engineered prompt

Build me a cross platform machine learning runtime that can run trained models quickly on different operating systems and hardware. It should let people load an ONNX model and use it for inference from common languages, and it should be set up so training support can be added for PyTorch style workflows too.

I want the project to feel production ready, with clear documentation, simple examples, and a clean way to build and test it on Windows, Linux, and macOS if possible. Make it fast by using hardware acceleration when available, and use sensible optimizations so it performs well on CPU and GPU. It should also include basic developer tooling, sample code, and a straightforward command line or library interface that shows how to run a model.

If you need current details or APIs, look up the latest docs online and follow the project standards already in the repo.