ubuntu GPU环境下使用MindSpore 1.0调试AI诗人的一次无疾而终的尝试

本文主要是介绍ubuntu GPU环境下使用MindSpore 1.0调试AI诗人的一次无疾而终的尝试，希望对大家解决编程问题提供一定的参考价值，需要的开发者们随着小编来一起学习吧！

转载地址：https://bbs.huaweicloud.com/forum/thread-83497-1-1.html

作者：longvoyage

电脑硬件配置：

12345

处理器英特尔 Core i5-7200U @ 2.50GHz 双核主板联想 LNVNB161216 ( 7th Generation Intel Processor Family I/O - 9D58 笔记本芯片组 ) 显卡 ‍Nvidia GeForce 940MX ( 2 GB / 联想 )‍ 内存 8 GB ( SK Hynix ) 主硬盘三星 MZNTN512HDJH-000L2 ( 512 GB / 固态硬盘 )

笔记本是win10的，也没多少空间了

手上刚好有个ssd固态U盘，做了一个wintogo 然后把ubuntu也装上去了

软件要求：

12345

Ubuntu 18.04 x86_64- Python 3.7.5 - CUDA 10.1 - CuDNN 7.6 - gmp 6.1.2

Ubuntu 18.04自带 python 3.6，需要再安装Python3.7.5

1. 安装Python 3.7.5

http://cdn.npm.taobao.org/dist/python/3.7.5/Python-3.7.5.tgz

修改Python源码包中ssl的参数

打开源码解压目录中的 Modules/Setup ,直接搜索 SSL= ，将SSL=后面的目录改为前面openssl的安装目录，并把下面三行的注释去掉。

此外必备的gcc make什么的要提取安装好

123456

tar -zxvf python3.7.5.tgzcd python3.7.5./configure --prefix=/usr/local/python375make && make installln -s /usr/local/python375/bin/python3.7 /usr/bin/pythonln -s /usr/local/python375/bin/pip3.7 /usr/bin/pip

2. 安装CUDA 10.1

https://developer.nvidia.com/compute/cuda/10.1/Prod/local_installers/cuda-repo-ubuntu1804-10-1-local-10.1.105-418.39_1.0-1_amd64.deb

1234	sudo dpkg -i cuda-repo-ubuntu1804-10-1-local-10.1.105-418.39_1.0-1_amd64.debsudo apt-key add /var/cuda-repo-<version>/7fa2af80.pubsudo apt-get updatesudo apt-get install cuda

安装好了配置环境变量

1234

export CUDA_HOME=/usr/local/cuda-10.1export LD_LIBRARY_PATH=/usr/local/cuda-10.1/lib64:/usr/local/cuda-10.1/extras/CUPTI/lib64:$LD_LIBRARY_PATHsexport PATH=/usr/local/cuda-10.1/bin:$PATHexport LD_LIBRARY_PATH="/usr/local/cuda-10.1/lib64:/usr/local/cuda-10.1/extras/CUPTI/lib64:/usr/local/cuda-10.1/lib64:/usr/local/cuda-10.1/lib64"

3. 安装CuDNN 7.6

https://developer.nvidia.com/rdp/cudnn-archive

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

123	sudo cp cuda/include/cudnn.h /usr/local/cuda/includesudo cp cuda/lib64/libcudnn* /usr/local/cuda/lib64sudo chmod a+r /usr/local/cuda/include/cudnn.h /usr/local/cuda/lib64/libcudnn*

4.安装gmp

安装依赖

sudo apt-get install m4

然后就是configure；make；make install

安装完了验证下安装是否成功

import numpy as np
from mindspore import Tensor
from mindspore.ops import functional as F
import mindspore.context as contextcontext.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))