Skip to content

Inference Templates ​

Feature Overview ​

ItemContent
Applicable RoleOperator
Navigation PathAI Infra(On-Prem) > Templates > Inference Templates
Page Route/powerone/fast-build-v2/inference-templates
Managed ObjectConfiguration, status, and relationships on Inference Templates

Beginner Explanation ​

An inference template is like an assembly list for a model service. It combines frameworks, specifications, default parameters, and visibility scope so users can quickly create services from the template during deployment.

Terms ​

TermDescription
TemplateDeployment plan selected when users create model instances.
Factor FormParameter set filled in by users when creating instances.
Dynamic ExpressionDynamically calculates field values or display conditions based on user input, model, precision, or resource conditions.

Confirm prerequisites for Inference templates, model scope, framework scope, specification recommendations, form parameters, and publication status, follow Main Operations, run Result Validation, and continue to the next page.

First-Time User Notes ​

Confirm that the task involves Configuration, status, and relationships on Inference Templates, and then follow the recommended order. If fields or state differ from expectations, check prerequisites before continuing downstream.

Prerequisites ​

  1. Model configuration, framework configuration, and VRAM estimation have been completed.
  2. Available resource specifications have been created and associated with target clusters.
  3. Image services and required storage capabilities have been connected.
  4. The parameters that users need to fill in when creating instances have been clarified.

Page Description ​

Use this page to manage inference templates, model and framework scopes, recommended specifications, form parameters, and publishing state.

Inference Templates

The page displays the inference template list, including template name, status, model scope, framework scope, update time, and operation entrypoints.

The following figure shows the inference templates page.

Main Operations ​

View Inference Templates ​

  1. Open the corresponding template-configuration page and filter by name, version, status, or update time.
  2. Open details and check associated models, frameworks, images, resource requirements, and current version.
  3. If no record is returned, reset filters. For incompatibility, first check dependencies.
  4. Redact internal images, storage locations, and startup configuration before sharing.

Create Inference Template ​

Pre-Operation Check ​

  1. Available frameworks, runtime images, and model configurations have been prepared.
  2. Resource specifications, deployment mode, and visibility scope adapted by the template have been confirmed.
  3. Default parameters have been confirmed not to expose internal paths, credentials, or test endpoints.
  4. Which user deployment entrypoints are affected by template changes has been clarified.

Procedure ​

  1. Go to AI Infrastructure > On-Prem > Templates > Inference Templates.
  2. In the top right, click "Create Inference Template".
  3. In the Model dropdown, select the target model (e.g., DeepSeek-R1-Distill-Qwen-7B).
  4. In the Model Version dropdown, select a version that does not have an inference template created yet.
  5. In the Framework Version dropdown, select an adapted inference framework version (e.g., vLLM · v0.13.0).
  6. In Template Name, verify or customize the template name (defaults to the model name).
  7. Confirm the model and framework dependencies, then click "Save" to finish creation.

Operation Screenshot ​

Create inference template

The image shows the Create Inference Template dialog. Select model, model version, and framework version, then confirm the template name. After saving, it serves as the standard deployment template for that model version.

Import or Export Inference Templates ​

Applicable Scenarios ​

Use the "Import/Export" menu to batch-maintain inference templates, or to export template configurations for audit, reconciliation, and controlled migration.

Steps ​

  1. Go to AI Infrastructure > On-Prem > Templates > Inference Templates.
  2. Click "Import/Export" and choose "Import" or "Export" according to the business purpose.
  3. For import, upload the file as required by the page and verify template, model, framework, resource specification, factor form, and visibility scope.
  4. For export, confirm the template filter scope, then generate and download the template configuration as prompted by the page.
  5. Before importing, verify that dependencies are available in the target environment. Save export files in a controlled directory.

Result Validation ​

  • After import, the Inference Templates list shows the added or updated template and state.
  • The template scope in the export file matches the current filter conditions.
  • Model, framework, specification, and parameter references in template details can be resolved correctly.

Notes ​

  • Template import depends on model, framework, image, and specification objects. Do not submit it when dependencies are missing in the target environment.
  • Import may update a template with the same identifier. Verify release scope, default parameter values, and downstream deployment impact first.

Parameter Quick Reference ​

Field NameRequiredField TypeExampleDescription
Template NameYesTextllm-inference-templateTemplate name users see when creating inference services. Use a maintainable name instead of temporary test naming.
DescriptionNoMulti-line textExample descriptionTemplate purpose, adapted model, and usage boundary. Write non-sensitive notes only. Do not include internal addresses or test parameters.
Applicable ScenarioConditionally requiredDropdown / enumModel inferenceBusiness scenario or model service type for the template. Keep it consistent with model type, framework capability, and user entry.
Publication ScopeConditionally requiredDropdown / enumExample valueScope where the template is published to the user side. Confirm affected tenants, regions, and entries before publishing.
Visibility ScopeYesDropdown / enumTenant AControls which users or tenants can use the template. Incorrect scope may make the template invisible or visible to non-target tenants.
ModelConditionally requiredDropdown / enumA100-SXM4-80GBModel or model set applicable to the template. Dependent objects must be available and match the framework.
Model VersionConditionally requiredDropdown / enumv0.8.0Model version referenced by the template. Keep it consistent with model path, quantization method, and VRAM estimation rules.
Model SourceConditionally requiredAddress / pathObject StorageModel file source, repository source, or object storage source. Do not write real model repository addresses, endpoints, or internal paths.
FrameworkYesNumber / capacityvLLMRuntime framework configuration called by the template. The framework must support the selected model type and runtime mode.
Framework VersionConditionally requiredTextv0.8.0Framework version referenced by the template. Confirm impact on existing templates and instances before modification.
Runtime ImageConditionally requiredText<BASE_URL>/<PROJECT>/<IMAGE>:<TAG>Container image used by the framework runtime. Confirm image region, registry permissions, and target cluster pull access.
Resource SpecificationYesNumber / capacitygpu-a100-1cardDefault recommended or selectable compute specification for the template. Match model VRAM, concurrency, context length, and deployment mode.
Deployment ModeYesDropdown / enumSingle instanceDetermines service replicas, scaling, and scheduling mode. Keep it consistent with framework startup commands and resource specifications.
VRAM Estimation RuleConditionally requiredNumber / capacity80VRAM rule used to recommend or validate specifications. Keep it consistent with model parameter count, quantization method, and context length.
Service PortConditionally requiredPort / number8000Actual listening or exposed port of the model service. Must match the actual framework listening port.
Port Exposure PolicyConditionally requiredPort / numberExample valuePort exposure method and authentication mechanism. Incorrect configuration may expand service exposure scope.
Port TagNoPort / numberOpenAI API PortIdentifies port protocol type or purpose. Keep it consistent with OpenAI API, Ollama API, or custom protocol.
Health CheckConditionally requiredAddress / path/healthPath or command used to determine whether the service starts successfully. Match the actual service path, port, and startup delay.
Factor FormConditionally requiredConfiguration textContext lengthParameter form that users fill in when creating instances. Verify fields, default values, and validation rules one by one.
Default ParametersNoDropdown / enum--max-model-len 8192Model or runtime parameters prefilled when creating services. Do not write real tokens, AK/SK, private keys, endpoints, or test values.
Dynamic ExpressionConditionally requiredDropdown / enum${gpuCount} >= 1Dynamically calculates fields or visibility based on model, framework, specification, or user input. Expression errors can cause missing fields or abnormal startup commands.
Trigger ConditionConditionally requiredDropdown / enumModel type = LLMControls fields displayed under specific models, frameworks, or options. Validate with actual model, framework, and specification combinations.
ActionsSystem-generatedAction entryEditAdd, save, submit, publish, OK, and similar page operations. Save, Submit, Publish, and OK are high-risk final actions.

Pitfalls ​

  • Publishing an inference template affects user-side selectable templates and the real service creation scope.
  • Incorrect visibility scope may make the template invisible or visible to non-target tenants.
  • Mismatched model, framework, image, specification, or VRAM rules can cause instance creation or startup failures.
  • Incorrect default parameters, dynamic expressions, or trigger conditions can cause missing user form fields, wrong parameters, or abnormal startup commands.
  • Incorrect port exposure policy may expand the service exposure scope.
  • Do not write real tokens, AK/SK, private keys, endpoints, internal addresses, model repository addresses, tenant IDs, cluster IDs, or test parameters.
  • Save, Submit, Publish, and OK are high-risk final actions. Confirm the scope and impact before executing the final action.

Result Validation ​

Check ItemSuccess SignalIf Abnormal
Page entryInference Templates opens with the target operation entryCheck Operator permission and whether the menu is available
Object recordConfiguration, status, and relationships on Inference Templates is visible in the list or detailsReset filters and verify name, ownership, and creation result
State resultState after creation or change matches the page messageCheck operation feedback, dependency state, and latest update time
Downstream useA downstream page can select or associate the targetReturn to prerequisites and check enabled state, ownership, and visibility

FAQ ​

An Imported Inference Template Cannot Be Published or Deployed ​

Symptom:

The template import completes, but its state is abnormal or the template cannot be used on later publish or deployment pages.

Possible Causes:

  • The model, framework, image, or specification is missing or unavailable in the target environment.
  • Factor fields, default values, or validation rules are incompatible.
  • Template visibility or state does not allow downstream use.

Solution:

  1. Open template details and verify the model, framework, image, and resource specification one by one.
  2. Open the corresponding configuration pages to verify dependency state and version compatibility.
  3. Check template visibility, release state, and factor-form validation rules.

Target Is Missing from Inference Templates ​

Symptom:

The page opens, but the expected Configuration, status, and relationships on Inference Templates is missing.

Possible Causes:

  • Filters remain active.
  • the object belongs to another scope.
  • a prerequisite is incomplete.

Solution:

  1. Reset filters
  2. verify region or tenant ownership
  3. confirm prerequisite state.

The Operation Entry on Inference Templates Is Unavailable ​

Symptom:

The create, register, or maintain entry is hidden or disabled.

Possible Causes:

  • Role permission is insufficient.
  • the page is read-only.
  • dependencies are not ready.

Solution:

  1. Check Operator permission
  2. read the page message
  3. complete dependency configuration first.

Inference Templates Has an Abnormal State After the Operation ​

Symptom:

A record exists after submission, but its state is unexpected.

Possible Causes:

  • Connectivity or validation failed.
  • a dependency is abnormal.
  • processing is incomplete.

Solution:

  1. Check feedback and update time
  2. inspect related objects
  3. troubleshoot the processing stage.

A Downstream Page Cannot Use Inference Templates ​

Symptom:

The current page is normal, but a downstream page cannot select or associate Configuration, status, and relationships on Inference Templates.

Possible Causes:

  • Visibility differs.
  • the object is disabled.
  • downstream cache is stale.

Solution:

  1. Check enabled state and ownership
  2. verify role visibility
  3. refresh and select again.

Notes ​

  • Template parameters must not contain real tokens, keys, or internal addresses.
  • Before publishing a template, confirm that dependent models, frameworks, images, specifications, and storage are all available.
  • Before modifying a published template, confirm the impact on user-side instance creation, template visibility scope, and active deployment entries.
  • Before save, submit, publish, or confirm actions, verify the scope and impact and execute them only after approval.

Next Steps ​

  1. Use a test tenant to create a model instance and verify the template.
  2. Adjust image, startup command, ports, and parameters based on failure logs.
  3. After template publication, periodically review model versions, framework versions, and specification scope.
  4. Review failure logs by model, framework, specification, and factor form combination to continuously calibrate the template.