Pytorch c++ Frontend pretrained model

Hello together,

I worked out this blog Applying Transfer Learning on Dogs vs Cats Dataset (ResNet18) using PyTorch C++ API :: Kushashwa Ravi Shrimali (Kush) — Hello Friend NG HZ Theme and i am little bit confused. In this blog the transfer learning is just a converted network with jit plus an additional layer. Is it possible to change the weights in the converted model? Or just i can change the weights of additional layers?

kind regards