Skip to content
ModAsteraDocumentation
日本語
1 of 3 · Deployments

Deploy and test a model

Create or open a deployment for an eligible completed model, activate it deliberately, and run a representative test.

Deployments overview with an inactive synthetic deployment, zero-usage summary, traffic filters, and chart headings
The overview shows deployment state and reporting controls. This synthetic example is inactive and has no traffic.

Check deployment eligibility#

  • A completed training label alone is not sufficient; the selected version must include a deployable model result.
  • If a deployment already represents the intended model, open and update that deployment instead of creating an accidental duplicate.
  • Activation controls whether the model accepts prediction traffic. Hiding a row does not deactivate its model.

Create and activate the deployment#

  1. Select the model version

    On the completed pipeline, choose the exact model version whose evaluation you reviewed.

  2. Select Deploy

    Confirm the deployment request once. If the platform finds an existing deployment for the intended model, open it rather than submitting another request.

  3. Wait for readiness

    Return to Deployments and wait until preparation finishes. Read the displayed state before retrying or changing the model.

  4. Review the deployment

    Confirm the name, description, selected model version or checkpoint, primary metric, and current active state.

  5. Activate deliberately

    Turn on Model is active only when the deployment is ready to receive traffic and affected API consumers are prepared.

Run a first UI test#

  1. Open the deployment

    Select the deployment name or its View action to open the detail page.

  2. Confirm model information

    Open model information and verify the task, input type, selected version, and expected output before uploading data.

  3. Provide one representative input

    Use the input control shown for the deployed task. Do not upload credentials, unrelated personal information, or production data merely to test the interface.

  4. Select Run Prediction

    Submit once and wait for the on-screen state to move from processing to a result or a clear error.

  5. Review the complete result

    Check the prediction summary, raw response, media or explanation panels, and the new execution-history entry.

Inactive synthetic deployment detail with model information, supported input types, and a disabled Run Prediction action
The detail page identifies the selected model and provides task-compatible input controls for browser testing.

Continue with

Test prediction inputs and explainabilityProvide the correct input for image, medical, tabular, or language models and inspect summaries, artifacts, explanations, and execution history.Manage and monitor deploymentsOperate deployment state and versions, review traffic and latency, and download a usage report for the selected period.Review pipeline results and model reportsInterpret task-specific evaluation visuals, distinguish operating points, compare versions, and create a model report grounded in the selected result.