Skip to content

Runtime Images ​

This task registers a cloud image used by inference frameworks. Image availability directly affects whether downstream services can start.

Target Outcome ​

The target cloud environment can pull the image, and inference-framework configuration can select it.

Applicable Roles ​

  • Platform Operator

Before You Start ​

  • Confirm the full repository address, controlled version tag, and supported framework types.
  • Confirm that the target cloud environment has network access and pull permission for the image repository.

Procedure ​

Add an Image ​

  1. Go to Deploy Assets > Runtime Images.
  2. Select Add Image.

Review runtime images

  1. Complete the form:
    • Image Name: maintain Simplified Chinese and English values separately.
    • Image Repository Address: enter the complete repository path, such as registry.example.com/ai/vllm.
    • Image Tag: enter a controlled version tag, such as 0.9.1-modelgallery.
    • Framework Type: select at least one compatible type, such as vllm, tgi, sglang, ollama, asr, tts, sdk-stable-diffusion, or comfyui.
    • Built In: select whether the image is provided as a platform built-in image.
    • Description: maintain optional localized usage descriptions.
  2. Select Save. Select Cancel to discard the form.

Add a runtime image

Parameters ​

FieldTypeExampleDescription
Image NameTextvllm in both language tabsRequired; maintain each language independently
Image Repository AddressTextregistry.example.com/ai/vllmRequired; complete image-repository path
Image TagText0.9.1-modelgalleryRequired; controlled image-version tag
Framework TypeMulti-selectvllm, tgi, sglangRequired; select at least one compatible type
Built InRadioYes / NoRequired; identifies whether the platform provides the image
DescriptionTextvllm 0.9.1 modelgalleryOptional localized description

Completion Checklist ​

Purpose: These are the exit criteria for this task. If any check fails, follow the troubleshooting section before continuing.

CheckPass Criteria
1Repository address and tag uniquely identify a controlled image.
2The target cloud environment can pull the image.
3Inference-framework configuration can select the image.

Troubleshooting ​

SymptomCheck First
Image pull failsRepository address, tag, pull permission, network, and architecture
Image pulls but the service fails to startImage content, framework compatibility, startup command, and port

User Manual ​

Review complete fields, validation rules, and troubleshooting for Runtime Images