How to profile using pure cpp torch?

Hi, I’m using torchlib to write a pure cpp program (without pytorch), how could I profile the GPU memory allocation information ? Thanks!

You can use Nsight Systems to profile the entire application.

Thanks for your advice!