We are planning migrate our project from python to C++. With Pytorch, as it include cuda/cudnn environment itself, so we can avoid unnecessery cuda configration work. However, with Libtorch, some tutorials show it need to install CUDA 10.0 otherwise the cmake can’t find the cuda library. Is it right? Or libtorch can be linked with python cudatoolkit 10.0.130 package?
When building the library, you need to install CUDA 10.0. As a user of your library, you don’t need to have CUDA installed, only the cudatoolkit 10.0.130 package is okay.