> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.umnai.com/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.umnai.com/_mcp/server.

# Serving

Deployment makes a trained model available for inference: using input values to produce predictions. A running deployment also supplies the outputs used for evaluation and explanations.

A trained model stores the result of learning. A deployment configuration defines its serving resources. A deployed model combines the two into the resource that accepts queries.

## How deployment works

#### Trained model

Start from an available model and confirm its dataset and training-job references.

#### Deployment configuration

Choose the serving type and resources appropriate to the workload.

#### Deployed model

Create the deployment from the model and configuration. Monitor it until its status is **RUNNING**.

#### Predictions and explanations

Use the running deployment in **Model Lab > Model Evaluation** and **Explanation Lab**.

## Deployment types

| Type in the app           | Use                                                                                 | Resource settings                        |
| ------------------------- | ----------------------------------------------------------------------------------- | ---------------------------------------- |
| **Serverless**            | On-demand workloads where scaling to zero is useful and cold starts are acceptable. | Memory size and maximum concurrency.     |
| **Real-time (Dedicated)** | Workloads that need the model to remain ready for low-latency requests.             | Serving machine type and instance count. |

Serving resources are separate from training and processing compute. Choose them for the model size and expected query workload; a training machine configuration cannot be substituted for a deployment configuration.

For models that use feature engineering, a running [Feature Transform Service](/ui-guides/deployment/feature-transform-services) applies the required input transformations. Select it during deployment and keep it running while the deployment uses it.

## Deployed model lifecycle

Under **Serving > Deployed Models**, inspect the deployment status and its model and configuration references. Use the deployed-model identity when evaluating or explaining predictions, so results can be traced to the intended trained model.

Deleting a deployment removes the serving resource and makes it unavailable for queries. Check whether it is still needed for evaluation or explanations before removing it.

## Next steps

Create a [deployment configuration](/ui-guides/deployment/configuration), then [deploy the model](/ui-guides/deployment/deployed-models).