All Products
Search
Document Center

Platform For AI:Job Template

Last Updated:Sep 11, 2026

In model training and tuning, even with the same code and data, different image versions, key environment variables, start commands, and resource specifications (such as GPU model and quantity) may lead to differences in training performance and results. DLC provides the Job Template feature: you can save validated training configurations as templates, preset and lock key items such as runtime images, resources, and mount paths, while only exposing necessary adjustable configurations such as hyperparameters and start commands. With job templates, you can standardize the base environment, reduce configuration errors, accumulate and reuse Best Practices, and help team members quickly create and reproduce experiments according to standards.

Usage Steps

Note

Workspace administrators and owners have full permissions for templates. Other workspace members have view and use permissions. Algorithm developers can only modify templates they created. Algorithm O&M engineers can modify all templates. For specific permission details, see Roles and permissions.

Create a Template

By saving common configurations as templates, you can achieve standardized task creation and avoid repetitive settings and manual errors.

  1. Enter template management. On the distributed training DLC page, click the Template button at the top to enter the template management page.

  2. Configure task pre-fill items. Click Create Template to enter the pre-fill page and set default values for tasks. This page is consistent with the task creation form, where you can fill in commonly used configurations as default values. You can also create a template based on an existing task by selecting a successfully run task as the pre-fill value.

    The Environment Information section on the pre-fill page includes the following configurable items:

    • Dataset Mount: Select a custom dataset and specify the mount path.

    • Storage Mount: Supports OSS, General-purpose NAS, Extreme NAS, CPFS, and CPFS for Smart Computing.

    • Startup Command: Shell editor.

    • Environment Variables: Supports multiple key-value pairs. Supports setting variables as Invisible to Others.

    • Third-party Library Configuration: Supports third-party library list and requirements.txt file directory.

  3. Configure task read-only rules. You can set pre-filled default values to read-only. Tasks created based on this template will be unable to modify read-only configuration items, preventing accidental modification of key parameters.

    The configuration page is divided into two sections. You can set each item to Read-only individually:

    • Environment Information: Image, Mount, Environment Variables, Code Configuration, Third-party Library Configuration, Startup Command

    • Resource Information: Resource Quota, Framework, Priority, Maximum Runtime

    Each section provides a Select All shortcut in the upper-right corner.

    Note

    For configuration items with parent-child hierarchy like environment variables:

    • If the parent directory is set to read-only, the subdirectories will also be set to read-only. When creating a task using the template, you cannot set newly added environment variables.

    • If the subdirectory is set to read-only, you can add and configure new environment variables when creating a task using the template.

      Read-only rules:

      When a subdirectory is set to read-only, the top-level Read-only check box is unchecked, and the Read-only check box for each environment variable (such as key1: value1, key2: value2) is independently selected.

      You can add and configure new environment variables when creating a task.

      Click + Add to add a new environment variable row, fill in the Key and Value, optionally check Invisible to Others to control visibility, and click the delete icon to remove the variable.

    It is recommended to set the Description field as required, which forces users to describe parameter adjustment information when creating tasks, making it easier to distinguish different tasks. You can define a naming convention in the template description (such as "research direction + adjusted parameters") to guide the team in filling it out consistently.

  4. Template preview. You can see that the environment variables set as read-only cannot be modified.

    The pre-fill configuration includes the following:

    • Dataset Mount: Custom dataset test_import_basic / v1, mount path /mnt/data/

    • Storage Mount: Supports adding OSS, General-purpose NAS, Extreme NAS, CPFS, and CPFS for Smart Computing.

    • Startup Command: Shell editor.

    • Environment Variables: key1:value1 and key2:value2

Create a Task Based on a Template

  1. After the template is created, go to the template list and click Create Task to create a task based on the template.

  2. On the task creation page, configuration items preset in the template will be auto-filled, and locked configurations (such as environment variables) will be displayed as read-only and cannot be modified. You only need to focus on the parameters that require adjustment and then create the task.

    The Environment Information section on the job creation page includes the following configuration items:

    • Startup Command: Shell editor.

    • Environment Variables: Key-value pairs preset by the template (such as key1:value1, key2:value2); once locked, they are displayed as read-only.

    • Third-party Library Configuration: Supports Third-party Library List or requirements.txt File Directory.

    • Code Configuration: Supports Online Configuration or Local Upload.

  3. Click the template name to enter the details page and view the Usage Record, which displays tasks created based on this template.

    The usage records page displays job information in a table, including Name/ID, Priority, Total Nodes, Job Resources, Username/ID, and Visibility scope. You can filter by time range and My Jobs Only. The Actions column provides Monitor, Clone, TensorBoard, Stop, and Resource Diagnosis buttons.

Note
  • Tasks created by cloning templates: Cloning a task created from a template will not be displayed in the template's usage records. When switching to the template tab, the task will not be automatically archived under the template.

  • Cloning vs. template usage recommendations: If you want to quickly launch tasks, flexibly modify configurations, and verify configuration correctness, use the clone feature. If you want to archive tasks in the same direction and be constrained by the template, it is recommended to create tasks based on templates.

Edit a Template

A created template can be Edit again.

In the template list, locate the target template and click Edit in the Actions column.

Clone a Template

A created template supports the Clone feature.

In the template list, locate the target template and click Clone in the Actions column.

Template Usage Tips

It is recommended to group experiments of the same research direction or similar types under a single template for centralized management. Through the template-task association, you can quickly view and locate corresponding tasks, improving experiment management efficiency.

On the Template tab, the template list displays the template name/ID, description, username/ID, creation time, modification time, and actions. You can expand a template to view associated job details (job name, framework, resource configuration, status, etc.). The Actions column supports Edit, Create Task, Clone, and Delete; sub-tasks support Monitor, TensorBoard, Stop, and Resource Diagnosis operations.