After building TensorFlow from source, seeing libcudart.so and libcudnn errors

Question

I\'m building TensorFlow from source code. The build appears to succeed; however, when my TensorFlow program invokes import tensorflow, one or both of the foll

Answer 1

On MacOS, this issue is often caused by bazel running in a sandbox environment, thus not respecting the LD_LIBRARY_PATH set in your local shell. I wouldn't bother going into the merit of deep integration of sandboxing in a build tool.

The simple workaround is to symlink the libraries into /usr/local/lib.

cd /usr/local/lib && ln -s ../cuda/lib/libcudart.8.0.dylib