When you create a dataset in Platform for AI (PAI), you can enable dataset acceleration for the dataset. When you create a Data Science Workshop (DSW) instance or submit a Deep Learning Containers (DLC) job, you can directly use an accelerated dataset to improve data reading efficiency. This topic describes how to use Dataset Accelerator in PAI.
Prerequisites
An accelerator is created. For more information, see Create and manage a dataset accelerator.
Enable dataset acceleration when you create a dataset
On the Datasets page, create a dataset and configure the parameters. The following table describes the parameters. For more information, see Create and manage datasets.
Parameter
Description
Create Dataset
Select From Alibaba Cloud.
Enable Dataset Acceleration
Select Enable Dataset Acceleration and configure relevant parameters to enable dataset acceleration.
Select an accelerator based on the selected data storage type, and configure the parameters of the slot, including the name, maximum capacity, and accelerated mount target. For more information, see Create and manage slots.
Click Submit.
The created dataset is displayed in the dataset list. The following figure shows the accelerated dataset.
Enable dataset acceleration for an existing dataset
On the Datasets page, click the name of the dataset to go to the Dataset Details page. For more information, see Create and manage datasets.
On the Dataset Details page, click Dataset Acceleration in the upper-right corner. In the Dataset Acceleration panel, select a dataset accelerator and configure the parameters of the slot. For more information, see Create and manage a slot.
Click Submit to enable acceleration for the dataset.
Use Dataset Accelerator
You can use Dataset Accelerator when you create DSW instances or submit DLC jobs.
When you create a DSW instance, you can select an accelerated dataset in the Storage section. For more information, see Create and manage a DSW instance.
When you submit a DLC job, you can select an accelerated dataset in the Datasets section. For more information, see Submit training jobs.
