Performance Difference between Python & C++ with Graph Optimization

[vackva]

I followed the origin graph optimization guide (https://onnxruntime.ai/docs/performance/graph-optimizations.html) and it seems that the optimization works great in Python (nearly cuts my interference time in half) but in c++ I don’t get any faster performance (loaded the optimized onnx model, tried different sessionSettings).

shouldnt the optimization behave the same across different languages? If so are there any lines of code missing in the tutorial?