Thank you for your question.
To answer your question better, could you please share more details about the steps and commands that you have used.
Here my steps:
conda create -n python3x -c intel python=3
source activate python3x
pip install --upgrade pip
pip install keras
conda install -c conda-forge bazel
conda install patch
pip install --upgrade setuptools
pip uninstall tensorflow
rm -rf tensorflow
git checkout r1.8 ## or r1.7 both same error
bazel clean ## used just in case of an error
bazel shutdown ## used just in case of an error
#all options with 'n' except kafka 'y'
#PATH = output
bazel build --config=mkl -c opt --copt=-mavx --copt=-mavx2 --copt=-mfma --copt=-mavx512f --copt=-mavx512pf --copt=-mavx512cd --copt=-mavx512er --copt="-DEIGEN_USE_VML" //tensorflow/tools/pip_package:build_pip_package
pip install --ignore-installed /home/uxxxxx/tensorflow_pkg/tensorflow-1.8.0-cp36-cp36m-linux_x86_64.whl
Are you referring here the same issue posted in the below thread created by you.
If yes, we can close this thread and continue in the previous one.
If not, could you please give more details on the following queries.
1. Are you executing the commands on login node or compute node.
2. Share the JAVA_HOME path
3. Share the model and commands you have used to run the same.
Thanks for the confirmation.
Hope you have successfully installed Intel optimized tensorflow 1.8.
To try out your issue, could you please share your model and required steps to run the same.
Hi, yeah We finally get the tensorflow_op file. but we still get the warnings and the error that I attached to this post. We don't know if we are missing something in the ./configure options. I selected all 'n' (jemalloc as malloc, google cloud, hadoop, Amazon, Kafka, XLA JIT, GDR, VERBS, OpenCL SYCL, CUDA, MPI, Clang)
I am also not sure about these PATH:
I executed which gcc:
Should I setup these PATHs as:
This is the model that we are running: https://github.com/tensorflow-northwest/kaggle/tree/master/projects/imaterialist_2018/DevCloud
I tried to build the .whl file again (same steps) and it finished well but when I ran the model, I'm getting this error:
libstdc++.so.6: version `CXXABI_1.3.8' not found
I already tried with --cxxopt="-D_GLIBCXX_USE_CXX11_ABI=0" when I execute `bazel build...`
Thank you for your patience!
myjob2.e81623.zip 1.0 K
Please check if you are getting the same when trying to import tensorflow from the python terminal.
If so, install libgcc and set the LD_LIBRARY_PATH.
1. conda install -c anaconda libgcc
2. export LD_LIBRARY_PATH=/home/uxxxx/.conda/envs/python3x/lib:$LD_LIBRARY_PATH
Hope this will help.
Could you please try in a new environment.
We are not getting the error that you have mentioned.
These are the steps to be followed in the compute node:
1. conda create -n python3x -c intel python=3
2.source activate python3x
3.pip install --upgrade pip
4. cd .conda/envs/python3x/lib/python3.6/site-packages
5.pip install --ignore-installed /home/uxxxx/tensorflow_pkg/tensorflow-1.8.0-cp36-cp36m-linux_x86_64.whl
6..conda install -c anaconda libgcc
8. From the python terminal import tensorflow and check whether you are getting the same error.
I tried with 4 different versions that I have done and I'm still getting the same error. When I run the model with the regular tensorflow it works slowly without any problem but when I install the .whl optimized I get the same errors described in this post.
I can import tensorflow as tf without a problem the error is when it tries to execute the line model.fit_generator(...
Thank you for your support!
Please try to add the environment activation (source activate python3x) step and LD_LIBRARY_PATH inside your 'im' script.
Kindly let us know if it helps.
We are not able to reproduce the error since the code is expecting some data in a particular folder.
Kindly share more information on the data that you have used to run the code so that we can try it from our end.