Other than the nightly builds, I want to be able to have libtorch tailored to my computer that isn’t quite set up like the one the nightly builds are made on. Cuda isn’t in /usr/local/cuda for example.
You can follow this doc:
https://raw.githubusercontent.com/pytorch/pytorch/master/docs/libtorch.rst
You need to create a build dir and run the commands there.
The paths can be set as arguments of tools/build_pytorch_libs.sh