Giters

prod-feng / Install-TensorFlow-on-CentOS-7-from-source

Install TensorFlow on CentOS 7 from source

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Install-TensorFlow-on-CentOS-7-from-source

Install TensorFlow on CentOS 7 from source

Summary: OS: CentOS 7.4.1708, 3.10.0-693.17.1.el7.x86_64

tensorflow-1.12

conda version : 4.6.4

conda-build version : 3.17.6

python version : 3.7.1.final.0

bazel 0.18.0

In the folowing I list the steps that I used to install TensorFlow on CentOS 7 from source.

  1. Install anacaonda3, and other needed packages There are a lot of online docs about how to install anaconda, so I skip the details of this step. One thing need to mention is: if possible, use conda to install the packages. I found that some pip packages are hard coded to be installed to /usr folder, which may cause trouble for HPC cluster users, who have no written previlege on that.

    conda install six wheel mock

  2. Install Bazel, which is the building tool used by TensorFlow.

    conda install bazel

    and other packages:

    pip install -U keras_applications==1.0.6 --no-deps

    pip install -U keras_preprocessing==1.0.5 --no-deps

  3. Download the source code of TensorFlow: (https://github.com/tensorflow/tensorflow/releases/tag/v1.12.0)

    tar zxf v1.12.0.tar.gz

    cd tensorflow-1.12.0

    ./configure

    bazel build --config=opmonolithic //tensorflow/tools/pip_package:build_pip_package

    There are some changes on the building settings may be needed to compile sucessfully. First, it seems TensorFlow does not compile on Python 3.7, because an issue in "protobuf" package. It is supposed to have been fixed in "protobuf" newer version of 3.6.1. The problem is that the bazel config file of TensorFlow, "tensorflow/workspace.bzl", is hard coded to use protobuf 3.6.0, so you may need to manually change it, like the following:

PROTOBUF_URLS = [

   "https://mirror.bazel.build/github.com/google/protobuf/archive/v3.6.1.3.tar.gz",
   "https://github.com/google/protobuf/archive/v3.6.1.3.tar.gz",
]
PROTOBUF_SHA256 = "73fdad358857e120fd0fa19e071a96e15c0f23bb25f85d3f7009abfd4f264a2a"
PROTOBUF_STRIP_PREFIX = "protobuf-3.6.1.3"

Do not forget update the "PROTOBUF_SHA256" key when change diffrent version of package.

  1. If everything goes well when compile, then you will be able to build a pip package of the TnesorFlow.

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

After it is done, you can then install the built TensorFlow package:

pip install /tmp/tensorflow_pkg//tensorflow-1.12.0-cp37-cp37m-linux_x86_64.whl

Great. It's done. You can verify to see the installed TensorFLow:

$ pip list|grep tensorflow

 tensorflow                         1.12.0

About

Install TensorFlow on CentOS 7 from source