All Products
Search

This Product

    Platform For AI:Preparations

    Document Center

    Platform For AI:Preparations

    Last Updated:Apr 15, 2026

    Prepare computing resources, a container image, a dataset, and source code for a DLC training job.

    Prerequisites

    If you use OSS for storage, grant DLC the required permissions to access OSS. Without proper permissions, I/O errors occur when accessing data from a mounted OSS bucket. For more information, see Cloud product dependencies and authorization: DLC.

    Step 1: Prepare resources

    Prepare compute resources for AI training. The following resource types are available:

    • Public resources

      Complete the DLC authorization. Public resources are then available on the Create Job page without adding a resource group.

    • General computing resources

      Create a dedicated resource group, purchase general computing resources, and create a resource quota to allocate the resources. Associate the resource quota with a workspace to submit training jobs. For more information, see General computing resource quotas.

    • Lingjun resources

      Prepare Lingjun resources and associate them with your workspace. For more information, see Create a resource quota.

    Step 2: Prepare an image

    Prepare a container image for the training environment. The following image options are supported:

    • Official image: PAI provides official images based on various frameworks. To view available images, go to the Images page of AI Asset Management in the PAI console. On the Image: page, on the Alibaba Cloud Images tab, set Modules to DLC to filter images that support DLC jobs.image

    • Custom image: If your training job requires specific environments or dependencies, use a custom image. Add the image as a PAI AI asset on the AI Asset Management > Images page of your workspace to reuse it across multiple training jobs. For more information, see Custom images.

      Important

      If you use a custom image with Lingjun resources, see RDMA: Use high-performance networks for distributed training for related considerations.

    • Image address: Specify the address of a custom or official image when submitting a training job. View image addresses on the Images page of AI Asset Management in the PAI console.

    Step 3: Prepare a dataset

    Upload training data to OSS, NAS, or CPFS and create a dataset, or directly mount data from an OSS bucket or a public dataset.

    Supported dataset types

    PAI supports datasets stored in OSS, General-purpose NAS, Extreme NAS, CPFS, and Lingjun CPFS. Dataset acceleration is supported for all types except Lingjun CPFS.

    Create a dataset

    For detailed steps, see Create and manage datasets. Note the following limitations:

    • OSS limitations: OSS is a distributed object storage service, not a file system. After mounting an OSS bucket, you cannot append data to or overwrite existing files.

    • CPFS VPC requirement: Configure the training job to use the same VPC as the CPFS file system. A VPC mismatch causes the job to remain in the Preparing environment state indefinitely.

    Enable dataset acceleration

    Enable dataset acceleration to improve data read efficiency. For details, see Use Dataset Accelerator in PAI.

    Step 4: Prepare source code

    Add your training code as an AI asset on the AI Asset Management > Code Configuration page of your workspace to reuse it across multiple training jobs. For more information, see Code Configuration.

    Next steps

    After completing these preparations, create a training job. For more information, see Create a training job.