What are the "correct" CMake settings to compile the PyTorch?

C++
1

Hi Everyone!
I would like to compile a custom version of the PyTorch. Could someone kindly tell me, what are the “safe” CMake settings that guarantee a successful compilation? In particular, we would like to use CUDA 12.1.
Thanks!
Piotr

2

You can just stick to the build instructions on GitHub, which should not use any custom cmake settings.

3

I use this and it works right will download torch automatically


set(Torch_DIR ${TORCH_INSTALL_DIR}/share/cmake/Torch)
Set(FETCHCONTENT_QUIET FALSE)
FetchContent_Declare(
  Torch
  URL       ${TORCH_URL}
  URL_HASH   MD5=${URL_HASH}
  SOURCE_DIR ${TORCH_INSTALL_DIR}
  OVERRIDE_FIND_PACKAGE
  DOWNLOAD_EXTRACT_TIMESTAMP
)
FetchContent_Populate(Torch)


find_package(Torch REQUIRED)