為什么從源碼編譯Tensorflow?
-
安裝過的人們都知道如果
pip install tensorflow
的話會報錯Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2
-
考慮到操作系統的的兼容性,現有的編譯好的二進制安裝包不適用openeuler,所以考慮這些因素,我們要自己搭建。
-
我們需要bazel交叉編譯軟件,有點類似cmake(上游下載的開源.sh文件已上傳到倉庫)bazel下載地址
-
chmod +x bazel-0.28.0-installer-linux-x86_64.sh
-
./bazel-0.28.0-installer-linux-x86_64.sh --user
-
bazel version
查看版本,configure.py查看tf所支持的版本(2.3版本所支持的bazel)
_TF_MIN_BAZEL_VERSION = '0.27.1'
_TF_MAX_BAZEL_VERSION = '0.29.1'
下載TF
git clone https://github.com/tensorflow/tensorflow
git barach -a
查看所有發行版本git checkout <version>
將分支同步到本地
開始編譯
-
bazel build --config=opt --config=cuda //tensorflow:libtensorflow_cc.so
(C文件庫) -
bazel build --config=opt //tensorflow/tools/pip_package:build_pip_package
(wheel文件) -
bazel build --config=opt --config=cuda //tensorflow:libtensorflow.so
(生成C庫) -
bazel build --config=opt //tensorflow/lite:libtensorflowlite.so
(Lite庫)
bazel編譯有的時候會出issue現bug,詳見issue
openeuler搭建Bulid Not Successfully 的Bug情況匯總
源碼編譯鏈接
官網上說安裝需要pip19以上
關於openeulerpip源默認指向python2該如何修改:
#!/usr/bin/python2=====>!/usr/bin/python3
# -*- coding: utf-8 -*-
import re
import sys
from pip._internal.cli.main import main
if __name__ == '__main__':
sys.argv[0] = re.sub(r'(-script\.pyw?|\.exe)?$', '', sys.argv[0])
sys.exit(main())
2、修改默認指向由於版本不同所發生的bug
from pip import main
ImportError: cannot import name 'main' from 'pip' (/usr/lib/python3.7/site-packages/pip/__init__.py)
source code change to this
from pip import __main__ //這行也要修改
if __name__ == '__main__':
sys.exit(__main__._main())//增加__main__._
PRODUCT
[root@openeuler lzb]# pip -V
pip 20.1.1 from /usr/local/lib/python3.7/site-packages/pip (python 3.7)
安裝 TensorFlow pip 軟件包依賴項(如果使用虛擬環境,請省略 --user 參數):
pip install -U --user pip six numpy wheel setuptools mock 'future>=0.17.1'
pip install -U --user keras_applications --no-deps
pip install -U --user keras_preprocessing --no-deps
setup.py 💯 we can know the vesion and dependence on REQUIRED_PACKAGES
> REQUIRED_PACKAGES = [
> 'absl-py >= 0.7.0',
> 'astunparse == 1.6.3',
> 'flatbuffers >= 1.12',
> 'gast == 0.3.3',
> 'google_pasta >= 0.1.8',
> 'h5py >= 2.10.0, < 2.11.0',
> 'keras_preprocessing >= 1.1.1, < 1.2',
> # TODO(mihaimaruseac): numpy 1.19.0 has ABI breakage
> # https://github.com/numpy/numpy/pull/15355
> 'numpy >= 1.16.0, < 1.19.0',
> 'opt_einsum >= 2.3.2',
> 'protobuf >= 3.9.2',
> 'tensorboard >= 2.3.0, < 3',
> 'tensorflow_estimator >= 2.3.0, < 2.4.0',
> 'termcolor >= 1.1.0',
> 'wrapt >= 1.11.1',
> 'w**h*eel >= 0.26',
> 'six >= 1.12.0',
> ]