ubuntu16.04 从源码安装tensorflow-gpu r1.4版本---2017-12-07

最新推荐文章于 2026-06-19 16:43:38 发布

原创最新推荐文章于 2026-06-19 16:43:38 发布 · 1k 阅读

1 ·

本内容遵循CC 4.0 BY-SA版权协议

收录于

tensorflow

本文详细介绍了如何从源码编译安装TensorFlow，并配置GPU支持。包括安装依赖库、配置环境变量、构建pip包及测试等步骤。

声明：以下文章基于https://www.cnblogs.com/lindaxin/p/7797477.html修改，感谢原作者

参考文章：https://www.tensorflow.org/install/install_sources，可以首先按照其中的说明安装依赖项，具体步骤参考下面

本文假设已安装好cuda 8.0和cudnn 6.0,安装参考文章：https://zhuanlan.zhihu.com/p/27890924


1、安装bazel
参考官网这里https://docs.bazel.build/versions/master/install-ubuntu.html#install-on-ubuntu,只有三步，很简单

2、安装第三方库

在terminal中输入以下命令，务必参考官方教程：https://www.tensorflow.org/install/install_sources

sudo
 apt-get install
 python-numpy swig python-dev python-wheel #安装第三方库

sudo
 apt-get install
 git

3、安装tensorflow

在terminal中输入以下命令，下载tensorflow源码

git
 clone https://github.com/tensorflow/tensorflow

在terminal中输入以下命令：执行配置文件，对安装步骤进行配置

cd
 ~/tensorflow #切换到tensorflow文件夹

./configure
 #执行configure文件

由于一直在更新，下面的配置步骤是2个月之前的，不一定完整，具体可搜索相关信息，一般遇到不太认识的就选n

Do
 you wish to use jemalloc as the
 malloc implementation? [Y/n] y

jemalloc
 enabled

Do
 you wish to build TensorFlow with Google Cloud Platform support? [y/N] n

No
 Google Cloud Platform support will be enabled for TensorFlow

Do
 you wish to build TensorFlow with Hadoop File System support? [y/N] n

No
 Hadoop File System support will be enabled for TensorFlow

Do
 you wish to build TensorFlow with the XLA just-in-time
 compiler (experimental)? [y/N] n

No
 XLA JIT support will be enabled for TensorFlow

Found
 possible Python library paths:

  /usr/lib/python2.7/site-packages

  /usr/lib64/python2.7/site-packages

Please
 input the desired Python library path to use.  Default is [/usr/lib/python2.7/site-packages]

 

Using
 python library path: /usr/lib/python2.7/site-packages

Do
 you wish to build TensorFlow with OpenCL support? [y/N] n

No
 OpenCL support will be enabled for TensorFlow

Do
 you wish to build TensorFlow with CUDA support? [y/N] y

CUDA
 support will be enabled for TensorFlow

Please
 specify which gcc should be used by nvcc as the
 host compiler. [Default is /usr/bin/gcc]:

Please
 specify the CUDA SDK version you want to use, e.g. 7.0. [Leave empty to use system default]:
 8.0

Please
 specify the location where CUDA
 8.0 toolkit is installed.
 Refer to README.md for more
 details. [Default is /usr/local/cuda]:
 /usr/local/cuda-8.0

4、创建pip,此步骤是从源码编译生成tensorflow的whl文件，可选

在tensorflow的根目录下，在terminal中输入以下命令：

bazel
 build -c opt //tensorflow/tools/pip_package:build_pip_package

bazel
 build -c opt --config=cuda //tensorflow/tools/pip_package:build_pip_package

bazel-bin/tensorflow/tools/pip_package/build_pip_package
 /tmp/tensorflow_pkg   可以在/tmp/tensorflow_pkg目录中找到生成的whl文件

5、从源码安装tensorflow环境

bazel
 build -c opt //tensorflow/tools/pip_package:build_pip_package

 #
 To build with GPU support:

bazel
 build -c opt --config=cuda //tensorflow/tools/pip_package:build_pip_package

mkdir
 _python_build

cd
 _python_build

ln
 -s ../bazel-bin/tensorflow/tools/pip_package/build_pip_package.runfiles/org_tensorflow/* .

ln
 -s ../tensorflow/tools/pip_package/* .

python
 setup.py develop

6、tensorflow测试

$
 python

 

>>>
 import tensorflow as tf

>>>
 hello = tf.constant('Hello,
 TensorFlow!')

>>>
 sess = tf.Session()

>>>
 print sess.run(hello)

Hello,
 TensorFlow!

>>>
 a = tf.constant(10)

>>>
 b = tf.constant(32)

>>>
 print sess.run(a+b)

42

>>>

大功告成

出现的错误
操作

bazel
 build -c opt --config=cuda //tensorflow/cc:tutorials_example_trainer

报错

ERROR:
 /home/yaroslavvb/tensorflow.git/tensorflow/tensorflow/core/kernels/BUILD:1080:1: undeclared inclusion(s) in rule '//tensorflow/core/kernels:cwise_op_gpu':

this is missing
 dependency dependency for following
 files included by 'tensorflow/core/kernels/cwise_op_gpu_floor.cu.cc':

  '/usr/local/cuda-8.0/include/cuda_runtime.h'

  '/usr/local/cuda-8.0/include/host_config.h'

  '/usr/local/cuda-8.0/include/builtin_types.h'

  '/usr/local/cuda-8.0/include/device_types.h'

  '/usr/local/cuda-8.0/include/host_defines.h'

  '/usr/local/cuda-8.0/include/driver_types.h'

  '/usr/local/cuda-8.0/include/surface_types.h'

  '/usr/local/cuda-8.0/include/texture_types.h'

可以进入tensorflow/third_party/gpus/crosstool/目录，打开CROSSTOOL文件，搜索cxx_builtin_include_directory，应该有三行，在下面添加行如下
cxx_builtin_include_directory: "/usr/local/cuda-8.0/include"

如果出现的错误是类似的，只要将cxx_builtin_include_directory: "/usr/local/cuda-8.0/include"的文件路径改一下就可以了，亲测有效

再次运行上一步的命令，应该就没问题了。

标签

#tensorflow #ubuntu16.04 #gpu