advice for simple onnxruntime script for ORTModelForVision2Seq (or separate encoder/decoder)

[eduard]

I am trying to use implement this class in C++ because unfortunately I didn't find any C++ implementation for this.

Therefore, my current approach is to revert this class and the auxiliary classes to a simple onnxruntime prediction, to make things easier to port to C++.

Does anyone have any advice in this matter? Thank you