All Products
Search

This Product

    Platform For AI:Manage dataset acceleration slots

    Document Center

    Platform For AI:Manage dataset acceleration slots

    Last Updated:Jun 20, 2026

    To use Dataset Accelerator, you must create an acceleration slot and associate it with a data source. The system pre-processes the dataset—based on its type, size, and the training model—to improve access speed. This topic describes how to create and manage acceleration slots.

    Prerequisites

    You have created a dataset acceleration instance. For more information, see Create and manage dataset acceleration instances.

    Create an acceleration slot

    1. Log on to the PAI console. In the left-side navigation pane, choose AI Acceleration > Dataset Accelerator.

    2. To open the Create Acceleration Slot panel:

      • Method 1: On the Acceleration Slots tab, click Create Acceleration Slot.

      • Method 2: From the instance details page.

        Note

        When using this method, the Instance parameter is automatically set to the current acceleration instance and cannot be changed.

        1. On the Acceleration Instances tab, click the name of the target instance to open its details page.

        2. On the Dataset Acceleration Slots tab, click Create Acceleration Slot.

    3. In the Create Acceleration Slot panel, configure the following key parameters.

      Parameter

      Description

      Instance

      Select an existing acceleration instance.

      Bind cloud product

      Object Storage (OSS)

      • Read/Write Attributes: Select Read-only or Read/Write.

      • Select OSS path: Select the OSS directory that contains the dataset.

      General-purpose NAS, Extreme NAS,

      • Select File System: Select the ID of an existing file system.

      • Mount Target: Select a mount target to access the file system.

      • File System Path: Select an existing storage path in NAS. Default: /.

      Maximum Capacity

      Set the capacity of the acceleration slot. The capacity must be greater than or equal to the dataset size. You can configure this parameter based on the size of the dataset that you want to accelerate.

      Mount Target

      This parameter is available only when Bind cloud product is set to Object Storage Service (OSS).

      Select or create a mount target to mount Dataset Accelerator on your training cluster. During model training, training jobs can access the acceleration slot through the mount target to read the associated dataset, improving data access efficiency.

      You can use one of the following methods:

      • Select Mount Point: Select an existing mount target.

      • Create Mount Target: Configure the following parameters to create a new mount target.

        • Mount Target Type: Only VPC is supported.

        • VPC Network: Select an existing Virtual Private Cloud (VPC). If no VPC is available, you can click create VPC to create one.

        • vSwitch: Select an existing vSwitch. If no vSwitch is available, you can click create vSwitch to create one.

      Note

      • Dataset Accelerator supports only VPC mount targets.

      • Training clusters in the same VPC can use a VPC mount target, even if the clusters are in different vSwitches.

      • When you add a mount target, the file system uses one IP address. We recommend that you select a vSwitch with enough available private IP addresses.

      • Only training clusters in the same VPC can mount the target.

      Enable automatic recycling

      Turn on the Enable automatic recycling switch to configure a runtime duration or a scheduled stop time for the acceleration slot.

    4. Click Submit.

      The system initializes the acceleration slot. You can use the slot when its Status changes to Running.

    Manage dataset acceleration slots

    You can manage dataset acceleration slots on the Dataset Accelerator page from the Acceleration Slots tab or from an acceleration instance details page. The available operations are the same on both pages. On the Acceleration Slots tab, you can view the slot list, including the slot name/ID, mount target, sync status, instance, accelerated storage, storage path, status, and load duration. In the Actions column, the More menu contains the Stop and Delete options.

    • Click a slot name to open its details page and view its Basic Information.

    • Hover over the Accelerated Storage column to view the dataset storage type.

    • You can stop or delete unused slots, or clone an existing one. You can also click Automatic Recycling to configure a runtime duration or a scheduled stop time.

    Manage mount targets

    To manage mount targets, click an acceleration slot's name to open its details page. Mount target configuration includes Mount target name, Mount target type (such as VPC), VPC ID, vSwitch ID, availability zone, IP, and port.

    • View configuration: Hover over a mount target name to view its configuration details.

    • View deployment configuration: Click a mount target name to display the deployment configuration for the Dataset Accelerator client.

    • Download deployment configuration: In the Deployment configuration panel, click Download to download the YAML file. Use this file to configure Dataset Accelerator in your training cluster.

      Alternatively, click the Copy button to copy the YAML content to your clipboard. The YAML file includes fields such as apiVersion: data.datasetacc.io/v1, kind: DataSetAccelerator, metadata (which contains the acceleration slot name), and spec (which contains frontEndAddr and frontEndPort).

    Next steps

    You can enable dataset acceleration when creating a Data Science Workshop (DSW) instance or submitting a Deep Learning Containers (DLC) job to speed up data access. For more information, see Use a Dataset Accelerator in PAI.