(article source | turbine cloud community focus/deep learning GPU cloud platform, artificial intelligence official experience url: gpushare.com/)

Hengyuan cloud in the history of the most complete platform using tutorial was born, help everyone AI training, run to win the school season!

Must see article

Users who use Hengyuan cloud for the first time can collect the following picture

There are only three steps before you can start your training

1. Upload local data to [Personal Data] (limited to 50G)

2. Go to the cloud market and choose GPU

3. Download [personal data/public data set] to the instance

▲ Focus!!

NO.1 Before uploading and downloading data, pack and compress data; otherwise, data cannot be transmitted. Upload the data to the compressed package in the instance.

NO.2. Save files in the instance to the specified directory/HY-tmp or /hy-nas.

Note:

/ HY-tmp is the fastest, but will be deleted 24 hours after shutdown, suitable for storing temporary training data. / HY-NAS is the fastest, limited to 50GB, suitable for storing common training data. For details, please refer to the data directory.

NO.3 Data upload and download can be performed in multiple modes, including tool transfer, command transfer, client tool such as FileZilla, command tool such as OSS command (official recommendation, fastest), and Sftp command.

Note:

* During Sftp connection configuration, the port and password for connecting [personal data] and the instance are different

* JupyterLab also supports upload and download, but applicable to small files, large files still please upload to [personal data], and then download to the instance

The following is the tutorial video. Before the demonstration, the data has been uploaded to “Personal Data” (limited to 50GB). The demonstration content is “Creating an Instance” and the files in “Personal Data” have been downloaded to the instance through Sftp command. B station is the official video link: www.bilibili.com/video/BV1Xf…

No.4 After creating an instance, you can install the required software package and save the environment by creating a Custom image. You are advised to repeat this step each time you install a new software package. You can view and use the custom image in My Image.

Note:

* New platform features [import image], support DockerHub and OSS import

* After a customized image is created, data in the /hy-nas directory can be migrated with the image/instance, but data in the /hy-tmp directory cannot be migrated

Based on article

【 The instance 】

☉ cloud market

You can browse the list of platform machines in “Cloud Market” to get a preliminary overview of the machine, including idle number, video memory, network, CPU, whether vouchers are available, whether there is shared storage, etc.

⊙ Creates instances.

When creating an instance, you can select the charging mode, GPU model number, instance mirroring, and automatic renewal. When placing an order, the voucher will be deducted first.

⊙ Instance Management

After an instance is created, you can configure it in Instance Management, including restart, shutdown (without charging), instance migration (without charging), instance initialization, custom image creation, and subcontracting week/month/year. Releasing an instance means deleting the instance.

【 data* * * *

⊙ Personal data

You can check the uploaded and downloaded files and storage usage in personal Data.

⊙ Public data set

The platform preset a large number of public data sets, including CV, NLP and other fields, platform users can use as required, the use method can refer to the relevant documents.

Senior post

[Terminal Login Example]

☉ Windows Login Instance

After obtaining the login information, you can use Xshell, PuTTY, MobaXterm and other SSH client instances to connect.

☉ Mac OS/Linux login instance

On The Mac OS/Linux operating system, you can log in using the built-in terminal application, such as iTerm2.

[Best Practices]

☉ Tmux

Tmux is a terminal multiplexer that can manage sessions and split screens. The Tmux tool has been installed in the official image, and you can use commands directly. For details, see the documentation tutorial.

☉ JupyterLab

JupyterLab is a new generation of Web-based Jupyter interactive development environment. You can use the Web to manage files, run Shell commands, and run Python codes. It contains all functions of the Jupyter Notebook.

☉ TensorBoard

TensorBoard is a tool for providing measurements and visualizations needed during machine learning workflows, as described in the documentation tutorial.

☉ VSCode

VSCode is a cross-platform code editor developed by Microsoft. It is free, open source, and supports extension plug-ins. Remote plug-in can be used to connect to a Remote server for development.

☉ PyCharm

PyCharm is a Cross-platform Python editor developed by JetBrains. Only the Professional version supports remote development. See the documentation for more details.

☉ Spyder

Spyder is a free, open source, cross-platform Python development environment that provides advanced code editing, interactive testing, debugging, and other features. For details, see the documentation.

☉ ikatago

The platform supports the remote use of GPU by ikatago dedicated to Go AI. For details, see the documentation.

Turbine wisdom at cloud website |gpushare.com