I think this question is well-posed but I’m not sure: if I define a model using LibTorch, is the dataflow graph still dynamically generated during execution just like as if I were writing python? Or does it actually get translated into some kind IR at some point?