The specific operation process of cloud GPU (Hengyuan cloud) training

简介

This blog mainly talks about how to use the cloud serverGPU跑程序,Mainly some settings and operation steps,The specific training steps can be found on my other blog.

The following steps may be missing,There may be some differences with you,If you have any questions, please comment or private message.

首先说一下环境,电脑操作系统是win10,The cloud server platform uses Hengyuan Cloud,IDE使用的Pycharm专业版（2022.1.3版,注意Community edition cannot use cloud server,Students can apply for the Professional Edition for free,具体步骤CSDNDetails on other blogs）,服务器操作系统是Linux.

云平台

Below is the link to Hengyuan Cloud,There are several main reasons for the choice：

1.Student certification will be available50元代金券,In addition, complete some novice tasks and recharge a small amount（大概30元）,There will also be some vouchers,See the novice task for details.

2.The help documentation is also more detailed,可以解决绝大部分问题.

3.GPUThe price is cheaper than other platforms,黄金会员有85折扣,Students bring their own95折扣.

4.You can get ingots by signing in,Used to redeem vouchers（The premise is that you have some money）

当然缺点也有一些

1.Platform changes will be more frequent,Basically a change in the direction of low welfare,For example, reducing the number of vouchers obtained、Reduce free free space for personal data（I just started using yes50G,现在为5G,But the fee is reasonable）等.

2.哈哈,Can't figure out the second point.Overall very good,After all, it has to be eaten properly,白嫖yyds,下面是my invite link,希望大家能支持一下,谢谢了.恒源云https://gpushare.com/auth/register?user=15*****5802&fromId=396d00d18df7&source=link

 ps There are many cloud server platforms,But in general the operation process is the same,Some differences can also be found through the official documentation、Customer service or find Du Niang to solve it.

云平台操作

Select the one you want to useGPU

 The picture above is the cloud market interface,The left side can be used for filteringGPU型号,Available on the rightGPU.It is divided into vouchers、High availability and activity,The latter two require cash,This demo uses the voucher area to demonstrate.

The picture above is the interface after clicking rent,Details are shown in the figure.需要注意的是,The mirror is the same howeverGPUNot at the same time will lead to training failure（我遇到过这个问题,Not particularly clear reasons,可能是镜像的问题）.

实例创建成功后,Open my instance and you can see the following interface,Here I will use the example I created before to show it,Some basic information has been marked on the figure.The next step is to start the instance,Boot through the instance management button,冲冲冲.

？？？问号脸（不搞了,没钱,到此结束）

看了一下,Suddenly it was only possible to pay with real money,Add another disadvantage,In fact, it is similar to the first disadvantage,Changes are too frequent,The vouchers can be used for a while and they can't be used for a while,Just re-rent one.

 

好了,The lease was successful,After startup, the interface is as shown above,接下来配置pycharm.

pycharm配置

打开设置,点击python解释器,Add new interpreters via the gear on the right（There are two ways to add servers,There is also a tool in the menu bar/Tools进行操作）

ssh -p 45269 [email protected]  端口号是45269,用户名root,主机是i-2.gpushare.com,After filling in, click OK to connect.

填写密码,And remember to save the password,方便下次连接.

服务器上的pythonThe interpreter is set by default,The remote path is modified as follows. 

注意：It is recommended to upload the project manually（下文有讲）,Automatic uploads may require a second modification of the remote path.

选择完成后点击确定即可,The next step is no brainer.

出现上图界面,即配置成功,点击应用并确定即可.

之后回到pycharm界面,You will see the image above in the bottom right corner,On the right is the configured remotepython解释器,On the left is the server,These two need to be in one-to-one correspondence,The server can be modified if it does not correspond,Click to edit to45269.

After the server is set up successfully,Updates to the interpreter and packages downloaded on the server are done in the background,If automatic upload is set,Automatic code uploads are also performed（If the dataset is in the project,will be uploaded together,It will cause the upload time to be very long,It is generally not recommended to upload data in this way,可通过oss上传,见下文）

 Automatic upload if not set,Manually upload files to the server by right-clicking on the project root directory,如上图.上传成功后,You can click through the Hengyuan Cloud web pagejupyterlab查看文件.

可以看到上传的文件,因为没有数据集,所以速度很快.

到此,pycharm配置基本完成.

数据上传

 Hengyuan cloud data upload is usedoss等多种方法,建议使用oss上传,首先下载oss到本地.

 

 下载完成后打开oss.exe,输入login进行登录,Enter the account and password of Hengyuan Cloud,登录成功

上传数据,Note that only compressed files can be uploaded.

 Upload via the command in the image abovec盘中的DOTADataset annotation files tooss中的DOTA文件夹,如下图.

Copy the dataset from the personal data directory to /hy-tmp目录下,通过在pycharm或者jupyterlabThe terminal implementation in .此处为jupyterlab的终端实现,首先进行登录,Then copy the file to hy-tmp目录,如下图.

 Then create a directory to store the unzipped filesmydata,and extract the dataset to mydata文件夹下.

 到此为止,pycharmAnd the server is all configured,After modifying the code and uploading it to the server,即可在pycharm中点击运行了（注意requirements.txtenvironmental requirements in ）.If the program reports an error, you can read my other blog,It may be because some settings or packages are not installed and so on.

YOLOv7使用云GPU训练自己的数据集https://blog.csdn.net/weixin_43764678/article/details/125863456?spm=1001.2014.3001.5501

 如果文章有错误,还请不吝赐教.