[WSL2 issues] WSL2-Ubuntu cmake configure error about pytorch-cpp

[mfl28]

For the first error ("Could not find compatible version") you probably point cmake to a libtorch version that is not yet currently compatible (i.e. tested) with the tutorials.

If you want to try a later libtorch version, you can modify the root-level CMakeLists.txt file to allow the later version.

For the second error: As mentioned in the Readme, if you chose to let the script create the scriptmodule files automatically, you must have python3 installed and Pytorch and torchvision (python versions!) must be importable by the python interpreter.

Hope this clears some things up for you!

[shaojunjie0912]

Thanks for your reply.

The first problem Could not find compatible Torch version (> = 1.12.0, < = 2.1.1): I use the cuda version of libtorch automatically downloaded so it has been solved .

The second question No module named 'torch': I found out that it was because CMake did not detect python in the conda virtual environment, but used the python that came with the system, so I change the code in CMakeLists.txt:

set if(CREATE_SCRIPTMODULES)
    find_package(Python3 COMPONENTS Interpreter REQUIRED)
endif()

to

if(CREATE_SCRIPTMODULES)
  set(Python3_EXECUTABLE "/home/olioolii/miniconda3/envs/pytorch/bin/python3")
endif()

Then everything works well.

Finally, thank you for your contribution to this project!

[mfl28]

Great! In this case you could probably even do:

cmake -B build ... -D CREATE_SCRIPTMODULES=ON -D Python3_EXECUTABLE=<path/to/python>

without modifying the cmake file.

[shaojunjie0912]

Got it! Thanks!