For some C++ headers needed by TFRA are not included in installation directory of old version Tensorflow. If you want to compile TFRA with these TF, it is necessary for extracting headers from TensorFlow compiling dirctory. Do as below:
- Compile TensorFlow with version you need to work with
-
Modify
BAZEL_CACHE_DIR
in ./env.sh, for$BAZEL_CACHE_DIR
is created randomly, so change it after step 1 :_BAZEL_CACHE_DIR=$BAZEL_CACHE_DIR/_bazel_root/860b4776177384b48cc611bdf8c7a91c
-
Extract headers by executing ./mkinclude.sh
./mkinclude.sh
- Copy
$OUTPUT_INCLUDE_DIR
to TFRA compile direcory
cp -r $_BAZEL_CACHE_DIR $TFRA_HOME/build_deps/tf_header/$TF_VERSION/
- Start compiling TFRA
PY_VERSION="3.7" TF_VERSION="1.15.2" TF_NEED_CUDA=0 sh .github/workflows/make_wheel_Linux.sh