GPU云服务器mindspore部署测试

1、准备Ubuntu18.04的ECS

2、制作私有镜像Ubuntu18.04

ECS虚拟机关机，生成私有镜像

3、使用私有镜像Ubuntu18.04开通GPU云服务器

登录ECS

4、mindspore环境准备

4.1 安装libsndfile包

sudo apt-get install  libsndfile*

4.2 root下执行：dpkg-reconfigure dash

在界面中选择no

4.3创建use1

#mkdir -p /home/user1

#useradd user1

#passwd user1

#chown -R user1.user1 /home/user1/

#cp .bashrc /home/user1/.bashrc

#chown user1.user1  /home/user1/.bashrc

4.4 user1用户安装Anaconda3

#wget https://mirrors.tuna.tsinghua.edu.cn/anaconda/archive/Anaconda3-5.3.1-Linux-x86_64.sh

#cp Anaconda3-5.3.1-Linux-x86_64.sh /home/user1/Anaconda3-5.3.1-Linux-x86_64.sh

# chown user1.user1 /home/user1/Anaconda3-5.3.1-Linux-x86_64.sh

切换用户安装Anaconda3

su - user1

$ bash Anaconda3-5.3.1-Linux-x86_64.sh

回车

按提示输入yes

提示信息“Do you wish to proceed with the installation of Microsoft VSCode? [yes|no]”，输入no；

4.5创建user1用户的python环境

vim ~/.bashrc

export PATH="/home/user1/anaconda3/bin:$PATH"

source ~/.bashrc

执行python命令确认python已安装

执行exit()退出python

conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/

conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/main/

conda config --set show_channel_urls yes

执行conda create -n seb python=3.7.5

切换到环境seb

source activate seb

4.6 安装cuda10.1

wget https://developer.nvidia.com/compute/cuda/10.1/Prod/local_installers/cuda_10.1.105_418.39_linux.run

#sudo sh cuda_10.1.105_418.39_linux.run

安装完成后检查确认

find / -type f -name libcuda.so.* -exec dirname {} \; 2>/dev/null

find / -type f -name libssl.so.* -exec dirname {} \; 2>/dev/null

如回显信息中有CUDA的版本信息证明CUDA安装成功

cd /usr/local/cuda-10.1/samples/1_Utilities/deviceQuery

make

./deviceQuery

user1执行

cd /usr/local/cuda-10.1/samples/1_Utilities/deviceQuery

./deviceQuery

user1用户执行

vim ~/.bashrc

export PATH=/usr/local/cuda-10.1/bin:$PATH

export LD_LIBRARY_PATH=/usr/local/cuda-10.1/lib64:$LD_LIBRARY_PATH

source ~/.bashrc

4.7安装cudnn

wget https://developer.nvidia.com/compute/machine-learning/cudnn/secure/v7.6.0.64/prod/10.1_20190516/cudnn-10.1-linux-x64-v7.6.0.64.tgz

#tar -zxvf cudnn-10.1-linux-x64-v7.6.4.38.tgz

#cd cuda

#cp ./include/* /usr/local/cuda/include

4.8安装GMP 

#sudo apt-get install m4

#wget https://gmplib.org/download/gmp/gmp-6.1.2.tar.bz2

#mkdir -p '/home/shims/install/gmp/share/info'

#tar -jxvf gmp-6.1.2.tar.bz2

#cd gmp-6.1.2

#./configure --prefix=/home/shims/install/gmp

#make

#make check

#make install

user1用户执行

vim ~/.bashrc

#gmp-4.3.2

export LD_LIBRARY_PATH=/home/shims/install/gmp4.3.2/lib:$LD_LIBRARY_PATH

source  ~/.bashrc

5、mindspore安装

#pip install --upgrade pip

#wget https://ms-release.obs.cn-north-4.myhuaweicloud.com/1.0.0/MindSpore/gpu/ubuntu_x86/cuda-10.1/mindspore_gpu-1.0.0-cp37-cp37m-linux_x86_64.whl

#cp mindspore_gpu-1.0.0-cp37-cp37m-linux_x86_64.whl /home/user1/mindspore_gpu-1.0.0-cp37-cp37m-linux_x86_64.whl

# chown user1.user1 /home/user1/mindspore_gpu-1.0.0-cp37-cp37m-linux_x86_64.whl

user1用户执行

source  ~/.bashrc

激活seb环境

source activate seb

安装mindspore

pip install --upgrade scipy==1.3.3 -i https://pypi.tuna.tsinghua.edu.cn/simple

pip install --upgrade sympy==1.7.1 -i https://pypi.tuna.tsinghua.edu.cn/simple

pip install mindspore_gpu-1.0.0-cp37-cp37m-linux_x86_64.whl

如报超时，加延迟检测时间参数：

pip --default-timeout=1000 install mindspore_gpu-1.0.0-cp37-cp37m-linux_x86_64.whl

pip show mindspore-gpu

6、mindspore测试

vim test.py

内容如下

import numpy as np

from mindspore import Tensor

from mindspore.ops import functional as F

import mindspore.context as context

context.set_context(device_target="GPU")

x = Tensor(np.ones([1,3,3,4]).astype(np.float32))

y = Tensor(np.ones([1,3,3,4]).astype(np.float32))

print(F.tensor_add(x, y))

执行python test.py查看结果