Tensorflow macOS binary

Tensorflow macOS binary, compiled with SSE4.1, SSE4.2 and AVX optimizations.

If you want to compile your own binary for macOS, check out compiling your own binary

Versioning

Latest available tensorflow version:

Getting Started

These instructions will get you a copy of the tensorflow running with SSE4.1, SSE4.2 and AVX enabled for optimizations.

Please DO NOT install the following binary if your CPU does not support the above optimizations.

How do I check for compatibility?

You should get the following warnings while running tensorflow which was downloaded through pip:

The TensorFlow library wasn't compiled to use SSE4.1/SSE4.2/AVX instructions, 
but these are available on your machine and could speed up CPU computations.

If this is not something that you see while executing tensorflow programs, please do not install this binary.

Installing

Download the binary from the git repository

git clone https://github.com/Thunderbottom/Tensorflow-macOS-binary.git

Open up your existing virtualenv:

source /path/to/virtualenv/bin/activate

Or create a new one:

# Make sure you have virtualenv installed
# or install it: pip install virtualenv (Use pip3 if pip is python2)
cd /path/to/whatever && virtualenv whatever

Install the downloaded wheel file with pip

# Use pip3 for python3
pip install filename.whl
# If you have an existing tensorflow installation, it'll be automatically removed and replaced.

Post installation

We need to check if tensorflow was properly installed. To check that, run the following command:

python3 -c 'import tensorflow as tf; print(tf.__version__)'

The output should look somewhat like this:

1.3.0

Built With

• Bazel
• Xcode CommandLineTools

Compiling your own binary

To compile your own version of tensorflow binary, you'll need the following:

python

brew install python python3

pip

# pip will probably be installed along with python
# if pip is not installed, you may go to
https://pip.pypa.io/en/stable/installing/

numpy1.12.1

pip install numpy

wheel

pip install wheel

bazel -

https://bazel.build/versions/master/docs/install-os-x.html

xcode commandlinetools -

# Run the following command in terminal
xcode-select install

Create a virtualenv -

# Make sure you have virtualenv installed
# or install it: pip install virtualenv (Use pip3 if pip is python2)
cd /path/to/whatever && virtualenv whatever

Open the virtualenv -

source /path/to/whatever/bin/activate

Clone the tensorflow repository -

# Replace <latest-version> with latest revision branch from repo
git clone https://github.com/tensorflow/tensorflow && cd tensorflow && git checkout <latest-version>

Run the configuration file and configure according to your needs -

./configure

After the configuration is complete, we will compile the binary -

# the following command will enable SSE4.1, SSE4.2 and AVX
bazel build -c opt --copt=-mavx --copt=-msse4.1 --copt=-msse4.2  -k //tensorflow/tools/pip_package:build_pip_package

If you wish to tune the flags, or add more compilation flags, you may do so by adding the following flags before the -k flag :

CUDA - --config=cuda

AVX2 - --copt=-mavx2

FMA - --copt=-mfma --copt=-mfpmath=both

After the compilation is complete, we will generate the wheel file -

bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp/tensorflow_pkg

Now, the wheel file will be generated and stored at /tmp/tensorflow_pkg

To install :

# Use pip3 for python3
pip install /tmp/tensorflow_pkg/filename.whl

Check out post-installation instructions to check the installed binary