Open greatblueheron opened 3 years ago
OK, so I did the following to make this work:
compile_opts += ' --compiler-options \'-fPIC -D_GLIBCXX_USE_CXX11_ABI=1\''
I suspect that the underlying issue with the original error was the gcc/g++ compiler version (tensorflow 1.15 can't stomach 8 or up), although I only got that explicit error message with the stylegan2 repo.
I get this with some of the downloaded weights but not all of them (eg lsun cats is fine, obama gives me the above error). I am using anaconda, with nvcc installed via conda install cudatoolkit-dev. Anyone else seeing this and found a workaround?