Ubuntu18.04 深度学习环境部署(cuda10.0 + pytorch1.1 + tensorflow2.0.0-beta + mxnet)

Ubuntu18.04 with deep learning (cuda10.0 + pytorch1.1 + tensorflow2.0.0-beta + mxnet)

Jul 30, 2020 | deep-learning-tools

1.install Ubuntu18.04 and update source

Update to mirrors.aliyun.com, reference this in step 1.

2.Deep learning environment

这里建议先安装驱动

优先采用第一种方法：

点开软件和更新

选择“Additional Drivers”,从下面的选项选择一个合适的版本　NVIDIA 1080Ti 选择410  Titan V 选435，　

第二种：命令行安装　NVIDIA driver

• sudo apt purge nvidia-* to remove old driver
• sudo add-apt-repository ppa:graphics-drivers/ppa to add repository
• sudo apt update && sudo apt upgrade
• ubuntu-drivers devices to see driver versions
• sudo apt install nvidia-driver-410 to install
• restart and check install nvidia-smi or watch -n 0.1 nvidia-smi
• use sudo apt-mark hold nvidia-driver-410 to prevent graphics drivers from updating automatically
  However, it may be slow and have some problem, you can download nvidia-driver-410.104 or other version, then install follow this reference to install.

Anaconda3.5.2-py36+cuda10.0+cudnn7.5+pytorch1.1+tensorflow2.0-beta+mxnet

Anaconda

• download Anaconda3-5.2.0-Linux-x86_64 or other version
• bash Anaconda3-5.2.0-Linux-x86_64.sh
• enter yes for adding path to .bashrc, no for install vscode
• source ~/.bashrc
• use which pip or which python, you will see default pip or python is now in Anaconda.
• use conda -V to version of Anaconda
• add conda mirror follow Anaconda 镜像使用帮助
• update pip source, mkdir ~/.pip && sudo gedit ~/.pip/pip.conf, then add the following in pip.conf

  [global] trusted-host=mirrors.aliyun.com index-url=https://mirrors.aliyun.com/pypi/simple/

NVIDIA driver

• sudo apt purge nvidia-* to remove old driver
• sudo add-apt-repository ppa:graphics-drivers/ppa to add repository
• sudo apt update && sudo apt upgrade
• ubuntu-drivers devices to see driver versions
• sudo apt install nvidia-driver-410 to install
• restart and check install nvidia-smi or watch -n 0.1 nvidia-smi
• use sudo apt-mark hold nvidia-driver-410 to prevent graphics drivers from updating automatically
  However, it may be slow and have some problem, you can download nvidia-driver-410.104 or other version, then install follow this reference to install.

CUDA

• download cuda_10.0.130_410.48_linux.run or search for other version
• sudo sh cuda_10.0.130_410.48_linux.run
• use q to install directly and pay attention not to install NVIDIA Accelerated Graphics Driver, the other’s can use y or default. Details are in reference.

• add following use sudo gedit ~/.bashrc

  export CUDA_HOME=/usr/local/cuda export PATH=$PATH:$CUDA_HOME/bin export LD_LIBRARY_PATH=/usr/local/cuda-10.0/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}

• source ~/.bashrc and use nvcc -V to see cuda version

cudnn

• download cudnn-10.0-linux-x64-v7.5.0.56.solitairetheme8 or other version
• then

  tar -zxvf cudnn-10.0-linux-x64-v7.5.0.56.solitairetheme8 sudo cp cuda/include/cudnn.h /usr/local/cuda/include/ sudo cp cuda/lib64/libcudnn* /usr/local/cuda/lib64/ sudo chmod a+r /usr/local/cuda/include/cudnn.h sudo chmod a+r /usr/local/cuda/lib64/libcudnn* cat /usr/local/cuda/include/cudnn.h | grep CUDNN_MAJOR -A 2

pytorch

• see https://pytorch.org/get-started/locally/

• or use downloaded .whl files

  pip install torch-1.1.0-cp36-cp36m-linux_x86_64.whl pip install torchvision-0.3.0-cp36-cp36m-linux_x86_64.whl

• test install

  In [1]: import torch In [2]: torch.__version__ Out[2]: '1.1.0' In [3]: torch.cuda.is_available() Out[3]: True

 

选装　tensorflow2.0-beta

• install use pip

  pip install tensorflow==2.0.0-beta0 # cpu pip install tensorflow-gpu==2.0.0-beta0 # gpu

• It may have problem ERROR: Cannot uninstall 'wrapt'. Directly use global search to remove related folders and .Egg-info related files. Details are in reference.

• test install

  In [1]: import tensorflow as tf In [2]: tf.__version__ Out[2]: '2.0.0-beta0' In [3]: tf.test.is_gpu_available() Out[3]: True

选装　mxnet

pip install mxnet-cu100

Applications(some are in reference)

sougou input

• download sougou for Linux
• Details are in reference.

WPS

• download WPS and install directly

minmize when click

gsettings set org.gnome.shell.extensions.dash-to-dock click-action 'minimize'

vim & Compression software & support exfat

sudo apt install vim sudo apt install p7zip-full p7zip-rar rar unzip sudo apt install exfat-fuse exfat-utils

git

• add your .ssh to your home and init

  sudo apt-get install git git config --global user.name '' git config --global user.email

opencv for python

sudo apt install libopencv-dev pip install opencv-python

pycharm

• install in Ubuntu Software
• or this reference, download pycharm Community version
• tar -zxvf pycharm-community-{your version}.tar.gz
• sudo mv pycharm-community-{your version} /opt/

• sudo gedit /usr/share/applications/pycharm.desktop and add the follows

  [Desktop Entry] Version=1.0 Type=Application Name=Pycharm Icon=/opt/pycharm-community-2019.1.3/bin/pycharm.png Exec=/opt/pycharm-community-2019.1.3/bin/pycharm.sh MimeType=application/x-py; Name[en_US]=pycharm

• sudo chmod u+x /usr/share/applications/pycharm.desktop

• It may have problem No module named 'distutils.core', solve it by
  sudo apt-get install python3-distutils