Wheel Specs
| TF | OS | Py | SSE3 | SSE4.1 | SSE4.2 | AVX | CPU |
|---|---|---|---|---|---|---|---|
| 1.14.1 | Ubuntu 18.04 | 3.7 | ✅ | ✅ | ✅ | ❌ | Intel(R) Atom(TM) CPU C2338 @ 1.74GHz (Silvermont) |
Download link
TBD (compiling in progress)
This page explains how to build from scratch Tensorflow 1.14.0 without AVX. From https://www.tensorflow.org/install/source and https://github.com/yaroslavvb/tensorflow-community-wheels
This package is built on :
- CPU Type:
Intel(R) Atom(TM) CPU C2338 @ 1.74GHz
To get you CPU Typegrep -m 1 'model name' /proc/cpuinfo - march:
silvermont
To get your marchgcc -march=native -Q --help=target|grep march
Your environment is on Ubuntu 18.04, and Python 3.6 or 3.7 is installed.
> python3 --version
Python 3.7.3We are starting by opening a screen because compiling Tensorflow will take ages. On this dual core Atom processor even after 6 hours it is still compiling. The screen will prevent interuption of the script when you will loose your shell.
screen -R tensorflowNote: if you lost your connection, connect again to your host then execute again screen -R tensorflow. You will get back to were you stopped
apt install build-essential python3.6-dev python3.7-dev
export tffolder="tensorflow-1.14.0"
mkdir ~/$tffolder && cd ~/$tffolder
python3 -m venv .venv
source .venv/bin/activate
pip install six numpy wheel setuptools mock future>=0.17.1
pip install keras_applications>=1.0.8 --no-deps
pip install keras_preprocessing>=1.0.5 --no-deps
pip install absl-py astor gast google_pasta numpy opt_einsum protobuf tensorboard tensorflow_estimator termcolor wraptFrom https://docs.bazel.build/versions/master/install-ubuntu.html#install-with-installer-ubuntu The snippet below is for the version 0.27.1, please update it with the current version you will find on their official repo here
Note from Bazel: Please downgrade your bazel installation to version 0.25.2 or lower to build TensorFlow! To downgrade: download the installer for the old version (from https://github.com/bazelbuild/bazel/releases) then run the installer.
apt-get install pkg-config zip g++ zlib1g-dev unzip wget
export bazelversion="0.25.2"
wget https://github.com/bazelbuild/bazel/releases/download/$bazelversion/bazel-$bazelversion-installer-linux-x86_64.sh
chmod +x bazel-$bazelversion-installer-linux-x86_64.sh
./bazel-$bazelversion-installer-linux-x86_64.shgit clone https://github.com/tensorflow/tensorflow.git
cd tensorflow
git checkout r1.14This script will help to select what flag you want
wget https://raw.githubusercontent.com/yaroslavvb/stuff/master/configure_tf.shexport flags="--config=opt --copt=-march=silvermont --copt=-mcx16 --copt=-mssse3 --copt=-msse4.1 --copt=-msse4.2 --copt=-mpopcnt --copt=-mno-avx -k"
export tag="silvermont"
export date="20190707"
./configure
bazel build $flags //tensorflow/tools/pip_package:build_pip_packageNote 1: Flags used are: -march=silvermont -mcx16 -mssse3 -msse4.1 -msse4.2 -mpopcnt -mno-avx
Note 2: Wait until 28,542 files are compiled. You can even take some vacations :D
./bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp/tensorflow_pkg
mkdir -p ~/$tffolder/tfbins/$date.$tag
cp `find /tmp/tensorflow_pkg -type f ` ~/$tffolder/tfbins/$date.$tag
bazel test $flags //tensorflow/...
bazel test $flags -j 1 //tensorflow/...
bazel build $flags //tensorflow/...export wheel=`find ~/$tffolder/tfbins/$date.$tag -type f`
export basename=`find ~/$tffolder/tfbins/$date.$tag -type f -printf "%f\n"`
cd ~/$tffolder/tfbins/$date.$tag
fullname=$date.$tag.$basename
ln -s $basename $fullnamepip install /tmp/tensorflow_pkg/tensorflow-version-tags.whldeactivatecd ~ && rm -R ~/$tffolderexit
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent