Closed klorinczi closed 3 years ago
Have you tried building with -march=native
?
It seems like you don't really have a nehalem CPU.
That would be -march=native
and none of the other -m
flags. The compiler will work it out.
This assumes you are compiling and running it on the same machine.
Also, BTW, if you use https://github.com/moodlehq/moodle-mlbackend-python/pull/31, you don't need tensorflow 1.14, and can use a supported version, installable via pip (though not apt-get).
Finally I was able to solve it by finishing the build from source correctly.
Just for reference for others, finishing install after building from source:
python3 -c 'import tensorflow'
pip3 uninstall tensorflow
./bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp/tensorflow_output
pip3 install /tmp/tensorflow_output/tensorflow-1.14.0-cp37-cp37m-linux_x86_64.whl
python3 -c 'import tensorflow'
Now the install was successful.
Unable to use Python machine learning backend (moodlemlbackend) v2.4.0 in Moodle v3.9.1.
I already had such problem in MDL-67193 with Moodlemlbackend v1.0.1 in Moodle v3.7.2. It was fixed successfully, but now I faced again and although I compiled tensorflow v1.14 from source, but still I get the error:
I also posted this report to Moodle Tracker (MDL-71523): https://tracker.moodle.org/browse/MDL-71523
I use
Installed with
sudo -H python3 -m pip install "moodlemlbackend==2.4.*"
I rebuilt tensorflow from source with following build options:
-march=nehalem -msse4.1 -msse4.2 -mpclmul -mpopcnt -maes -mno-avx -mno-avx2
But still I get the error:
moodlemlbackend requirements satisfied:
Any idea how can I make it work? How could I debug the problem?