> 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.

# Deployment configuration

Open **Serving > Deployed Models > Configurations**. A deployment configuration defines the serving resources for a model. Save the configuration first, then select it when creating the deployment.

## Before you begin

Identify the trained model and expected serving workload. Decide whether you need on-demand serverless serving or dedicated instances. Training and processing compute configurations are separate resources and cannot be substituted for a deployment configuration.

## Create a deployment configuration

1. Select **Create** and enter a **Configuration name**.

2. Select a **Deployment type**. The example uses **Serverless**, which provides on-demand scaling. For dedicated serving resources, select **Real-time (Dedicated)**.

3. For Serverless, choose **Memory size (MB)**. For Real-time (Dedicated), choose the serving **Machine type**.

4. For Serverless, set **Maximum concurrency**. For Real-time (Dedicated), set **Instance count** to at least **1**.

5. Select **Create**, then open the saved configuration and check the serving settings.

The Serverless example uses **6144 MB** of memory and maximum concurrency **1**. The memory choices range from **1024 MB** to **6144 MB** in **1024 MB** increments. Choose memory and concurrency for the model size and expected request volume; the example values do not establish capacity for your workload.

Give each configuration a name that distinguishes its serving setup. When creating a deployment, verify that you have selected the configuration with the intended deployment type and resources.

![Serverless configuration example with 6144 MB memory and concurrency 1.](/_fern-files/_img/adf9a1b4d49743830c8e5d3a1e88ff9b032e2962b871dd9d8c19901f2c65a8a4.webp)

## Choose serving resources

| Control             | Applies to            | What it determines                                                                     |
| ------------------- | --------------------- | -------------------------------------------------------------------------------------- |
| Configuration name  | Both types            | The reusable serving setup selected when deploying a model.                            |
| Memory size (MB)    | Serverless            | Memory allocated to the serverless workload. Choose from the values shown in the form. |
| Maximum concurrency | Serverless            | Maximum concurrent invocations. Review it alongside expected request volume.           |
| Machine type        | Real-time (Dedicated) | The serving machine, with its CPU and memory resources.                                |
| Instance count      | Real-time (Dedicated) | Number of serving instances; enter at least 1.                                         |

The current Serverless form starts at **2048 MB** and concurrency **100**. The screenshot shows a deliberately different example. Neither the starting values nor the example establish capacity for a particular model; verify the saved settings and evaluate the deployment with your workload.

## Review and reuse a configuration

Open the saved record and confirm the deployment type and the settings that apply to that type. Reuse the configuration when another deployment needs the same serving setup. Create a separate configuration when testing a different machine, memory allocation or concurrency limit so that the selected setup is unambiguous.

## Next steps

Continue to [Deploy models](/ui-guides/deployment/deployed-models).