You need to enable JavaScript to run this app.
Lake AI Service

Lake AI Service

Copy page
Download PDF
Daft
LAS job development
Copy page
Download PDF
LAS job development

This article describes how to use the development machine for job development.
The typical approach for job development is to start a development machine for remote job development. After job development is complete, package the job and use the LAS task management feature to run the job online. Users can also choose to develop jobs locally, but due to issues such as operating system compatibility, model support, and Python package support, it is recommended to use the LAS pre-configured development machine for job development.

Prerequisites

tip

It is recommended to mount file storage vePFS or object storage TOS when creating a development machine. The reasons are as follows:

  • The development machine may need to be rebuilt; using vePFS/TOS allows code to be preserved.
  • The development machine can share vePFS/TOS with the containers for distributed tasks, allowing jobs to run without packaging images or job packages (details will be explained below).
  • ve-daft has been installed and updated to the latest version. When creating a development machine, if the ve-daft image is selected, ve-daft and its Python dependencies and model dependencies are all installed.
    The method for updating ve-daft within the development machine is as follows:

    1. Click to download the update package.
    2. Install ve-daft and the operator package provided by LAS.
    # Install ve-daft and the operator package provided by LAS
    pip install {Installation package name}[las]
    pip install {Installation package name}[flash-attn] # Model acceleration tools
    
  • The development machine has been connected remotely via SSH. For detailed instructions, refer to: Remote connection to the development machine.

Job development and online execution

Job development

Method 1: Use vePFS/TOS mounting

Image

Method 2: Use image

Image

tip

Method two has obvious drawbacks during the development phase due to the need to frequently build and download images, but it offers significant advantages at runtime by ensuring environment stability through images.

Code development

Once the environment is ready, code development can be performed in TRAE/vscode/terminal.

Online execution

Refer to Task management to submit the developed job to task management for execution.

Method 1: Use vePFS/TOS mounting

Image

  1. Fill in the base image that matches the development machine, or at least a newer image from the same series.
  2. Enter the command to invoke the main Python script. It is recommended to use absolute paths for Python commands whenever possible.
  3. Enter environment variables. Common environment variables are usually as follows.
export ACCESS_KEY=xxx
export SECRET_KEY=xxx
export TOS_ENDPOINT=https://tos-cn-beijing.volces.com  # For intranet, use https://tos-cn-beijing.ivolces.com
export TOS_REGION=cn-beijing
  1. The mounted vePFS/TOS path. Ensure the mount path is consistent with the development machine.

Method 2: Use image

  1. First, save the development machine image on the development machine page.
  2. As in Method 1, select the newly created image on the task management page, fill in the startup command and environment variables (no need to mount vePFS/TOS).
Last updated: 2026.05.24 15:25:27