Tensorflow：安装和配置

骑猪看日落 2022-06-11 23:49 327阅读 0赞

# 安装anaconda #

windows下直接安装，最好勾上python加入系统路径。

**查看及安装tensorflow依赖**

查看cuda版本（机器支持gpu的不考虑）  
cat /usr/local/cuda/version.txt

cuda版本及其对应tf版本：

\[[经过测试的构建配置][Link 1]\]

查看python版本

python -V

安装python环境anaconda参考\[[python虚拟环境安装和配置][python]\]

# 安装对应的tf版本 #

### **检测已安装版本** ###

$pip3 show tensorflow

$pip3 show tensorflow-gpu

Note:  
1** 安装的tensorflow版本要和python对应**，否则出错：tensorflow\_gpu-1.12.0-cp...m-linux\_x86\_64.whl is not a supported wheel on this platform。

2 安装的tensorflow版本也要和cuda版本相对应，否则出错：tenforflow ImportError: /lib64/libstdc++.so.6: version \`CXXABI\_1.3.8' not found stackoverflow。

3 tensorflow\_gpu版本太高，如1.12，会报错Failed to get convolution algorithm. cudnn failed to initialize…。目前版本的TensorFlow的ObjectDection中，使用tensorflow-gpu ==1.12 版本会报错，如上，将版本降到tensorflow-gpu ==1.9.0即可正常运行。pip3 install --upgrade --force-reinstall tensorflow-gpu==1.10.1

4 tensorflow-gpu版本太低，如1.9，也会报错：tensorflow-gpu 1.9.0 module 'tensorflow.contrib.estimator' has no attribute 'stop\_if\_no\_increase\_hook'，这个模块在1.10以上版本才有，所以需要权衡一下再安装。

\[一些不同平台不同版本地址[https://stackoverflow.com/questions/55877398/how-to-install-tensorflow-version-1-12-0-with-pip][https_stackoverflow.com_questions_55877398_how-to-install-tensorflow-version-1-12-0-with-pip]\]

## linux中pip直接安装 ##

**安装最新版本：**

pip install tensorflow

**安装指定版本2.\***

pip install tensorflow==2.\*

**安装带gpu的指定版本1.\***

pip3 install --upgrade --force-reinstall tensorflow-gpu==1.10.1

pip3 install https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow\_gpu-1.12.0-cp36-cp36m-linux\_x86\_64.whl  
低版本下载地址在这里\[[https://pypi.org/project/tensorflow-gpu/1.12.2/\#files][https_pypi.org_project_tensorflow-gpu_1.12.2_files]\]

\[[安装 TensorFlow][TensorFlow]\]

## mac下安装 ##

**mac intel cpu安装**

pip3 install --upgrade https://download.tensorflow.google.cn/mac/cpu/tensorflow-1.13.1-py3-none-any.whl  
pip3 install --upgrade https://storage.googleapis.com/tensorflow/mac/cpu/tensorflow-1.12.0-py3-none-any.whl

tf1.12这种低版本的需要python3版本低，比如py3.6，否则import tensorflow时会提示出错：todo...

**mac m1芯片安装**

m1芯片安装就有点麻烦了，而且目前lz只找到tf2.0的版本，tf1.0貌似无法安装，有方法求留言！

m1芯片直接安装tf1.0会提示错误：[“zsh: illegal hardware instruction python” ][zsh_ illegal hardware instruction python]；

m1芯片直接安装tf2.0可能会出错：Process finished with exit code 132( signal 4 (SIGILL)。可能是因为“tensoflow版本问题，因为1.5.0以上的tensorflow需要AVX支持，所以，退回到1.5.0可以解决此问题。”

**需要使用特定版本的tf:**

**下载anaconda arm**  
打开anaconda arm 链接[https://conda-forge.org/blog/posts/2020-10-29-macos-arm64/][https_conda-forge.org_blog_posts_2020-10-29-macos-arm64]，点开here就可以下载了。

安装anaconda：/bin/bash ./Miniforge3-MacOSX-arm64.sh

conda create -n tfenv python=3.8

创建虚拟环境

cd  
conda create -n tfenv python=3.8  
激活虚拟环境  
conda activate tfenv

查看当前python路径  
which python

**下载tensorflow**  
[https://github.com/apple/tensorflow\_macos/releases][https_github.com_apple_tensorflow_macos_releases]  
进入这个网站，有Assets，点击下载文件tensorflow\_macos-0.1alpha3.tar.gz就可以了。

**在当前tfenv环境中执行下面的语句**  
libs="/Users/pipi/download/tensorflow\_macos/arm64"

env="/Users/pipi/miniforge3/envs/tfenv"

pip install --upgrade pip wheel setuptools cached-property six

pip install --upgrade -t "$env/lib/python3.8/site-packages/" --no-dependencies --force "$libs/grpcio-1.33.2-cp38-cp38-macosx\_11\_0\_arm64.whl"

pip install --upgrade -t "$env/lib/python3.8/site-packages/" --no-dependencies --force "$libs/h5py-2.10.0-cp38-cp38-macosx\_11\_0\_arm64.whl"

pip install --upgrade -t "$env/lib/python3.8/site-packages/" --no-dependencies --force "$libs/numpy-1.18.5-cp38-cp38-macosx\_11\_0\_arm64.whl"

pip install --upgrade -t "$env/lib/python3.8/site-packages/" --no-dependencies --force "$libs/tensorflow\_addons\_macos-0.1a3-cp38-cp38-macosx\_11\_0\_arm64.whl"

pip install absl-py astunparse flatbuffers gast google\_pasta keras\_preprocessing opt\_einsum protobuf tensorflow\_estimator termcolor typing\_extensions wrapt wheel tensorboard typeguard

pip install --upgrade -t "$env/lib/python3.8/site-packages/" --no-dependencies --force "$libs/tensorflow\_macos-0.1a3-cp38-cp38-macosx\_11\_0\_arm64.whl"

验证：

python

>import tensorflow

\[[Mac m1芯片anaconda上安装tensorflow][Mac m1_anaconda_tensorflow]\]\[[Macbook m1芯片版本怎么安装tensorflow2.0][Macbook m1_tensorflow2.0]\]。

之后可能还要安装其它包，如pandas，import时出错：ValueError: numpy.ndarray size changed, may indicate binary incompatibility. Expected 88 from C header, got 80 from PyObject

解决：重装numpy: pip uninstall numpy    pip install numpy

## window下安装 ##

安装老版本

pip3 install --upgrade https://storage.googleapis.com/tensorflow/windows/cpu/tensorflow-1.12.0-cp36-cp36m-win\_amd64.whl

### 安装tf相关拓展包 ###

pip3 install git+https://github.com/guillaumegenthial/tf\_metrics.git

python  
import tensorflow as tf

tf.\_\_version\_\_ \#查看tensorflow版本

tf.\_\_path\_\_ \#查询tensorflow安装路径

from: [\-柚子皮-][-_-]

ref:

[Link 1]: https://www.tensorflow.org/install/source#tested_build_configurations
[python]: https://blog.csdn.net/pipisorry/article/details/39998317
[https_stackoverflow.com_questions_55877398_how-to-install-tensorflow-version-1-12-0-with-pip]: https://stackoverflow.com/questions/55877398/how-to-install-tensorflow-version-1-12-0-with-pip
[https_pypi.org_project_tensorflow-gpu_1.12.2_files]: https://pypi.org/project/tensorflow-gpu/1.12.2/#files
[TensorFlow]: https://www.tensorflow.org/install/pip
[zsh_ illegal hardware instruction python]: https://stackoverflow.com/questions/65383338/zsh-illegal-hardware-instruction-python-when-installing-tensorflow-on-macbook
[https_conda-forge.org_blog_posts_2020-10-29-macos-arm64]: https://conda-forge.org/blog/posts/2020-10-29-macos-arm64/
[https_github.com_apple_tensorflow_macos_releases]: https://github.com/apple/tensorflow_macos/releases
[Mac m1_anaconda_tensorflow]: https://blog.csdn.net/lyang1896/article/details/113187799?utm_medium=distribute.pc_relevant.none-task-blog-2~default~BlogCommendFromMachineLearnPai2~default-1.control&depth_1-utm_source=distribute.pc_relevant.none-task-blog-2~default~BlogCommendFromMachineLearnPai2~default-1.control
[Macbook m1_tensorflow2.0]: https://blog.csdn.net/knowledge_endlesss/article/details/113853831
[-_-]: https://me.csdn.net/pipisorry