ImportError when import _openmmnn

Roy-Kid commented 3 years ago

hi I get an error when from openmmnn import * :

ImportError: /home/roy/tensorflow/lib/libtensorflow.so.2: undefined symbol: _ZNK10tensorflow6Device19IsRemoteCallAllowedEv

It quite seems like some issues aforementioned, but I have tried those solutions but still wrong with it. Do you know how to fix it?

here is how I install the plugin:

rm -rf build

mkdir build
cd build

export TENSORFLOW_DIR=/home/roy/tensorflow

export LD_LIBRARY_PATH=${LD_LIBRARY_PATH}:${TENSORFLOW_DIR}/lib
export CPLUS_INCLUDE_PATH=${CPLUS_INCLUDE_PATH}:${TENSORFLOW_DIR}/include

condaEnv=~/miniconda3/envs/admp
CUDA_DIR=/local/CUDA/9.0.176
export OPENMM_CUDA_COMPILER=$CUDA_DIR/bin/nvcc

cmake .. -DOPENMM_DIR=$condaEnv \
      -DCMAKE_INSTALL_PREFIX=$condaEnv \
      -DTENSORFLOW_DIR=$TENSORFLOW_DIR
      -DCUDA_TOOLKIT_ROOT_DIR=$CUDA_DIR \
      -DNN_BUILD_CUDA_LIB=1 \
      -DCMAKE_CXX_FLAGS='-std=c++11' \
      -DCMAKE_CXX_FLAGS:STRING=-D_GLIBCXX_USE_CXX11_ABI=0 
make -j4
make install
make PythonInstall

and the openmm is install directly from conda with v7.6 Thanks in advanced!

peastman commented 3 years ago

The most likely explanation is that you're compiling against one version of TensorFlow but linking to a different version at runtime. Notice that the error is happening inside libtensorflow.so. That is, the TensorFlow library is getting an error trying to find a TensorFlow function! Make sure you don't have any other version of TensorFlow in your library path.

virtualmartire commented 7 months ago

@Roy-Kid did you solved the problem? I am having it too

openmm / openmm-tensorflow

ImportError when import _openmmnn #17