oonxruntime-gpu: Reduce overall project size by levearging oonx cpu and gpu runtimes

[arky]

The current overall deployment of AddaxAI requires 4GB install file and 20 GB of free disk space. The most of the space is taken by python environments.

Python projects are now leverage on oonxruntime (https://pypi.org/project/onnxruntime-gpu/) packages to directly models. This issue is filed to explore the possibility of exploring this new feature to reduce the footprint of AddaxAI

[PetervanLunteren]

Thanks for sharing @arky! Do I understand correctly that onnxruntime-gpu only works for ONNX format models?

[arky]

@PetervanLunteren Yes, onnxruntime-[cpu,gpu] is specific to Microsoft ONNX format.