tensorflow build - CPU / MKL / Windows

问题:

Has anyone succeeded to build tensorflow python wheel with the following configuration:

1. CPU (not GPU)
2. OS: Windows 7 / server 2012
3. Using Intel MKL and/or mkl-dnn
4. Python 3.6

I am struggling, for days now, trying to tweak bazel files and cmake files without success.

Would like to know if someone succeeded and willing to share what he did.

Thanks, Lior

回答1:

Solution: 1

Installing TensorFlow from Source on windows is not generally supported. But you can try building this using bazel or tensorflow cmake build. Refer: https://www.tensorflow.org/install/install_sources

Solution: 2

Create Intel optimized conda environment as below:

conda create -n tf -c intel python=3.6

To install tensorflow:

conda install -c intel tensorflow

This will install tensorflow version 1.2.1 with mkl. This might help.

Prerequisites: Install Anaconda on your windows machine.

回答2:

I used this distribution and it worked fine for me (Windows 7 and 10 64bit, 2x Intel Xeon E5-2637 v3)

https://github.com/fo40225/tensorflow-windows-wheel