> 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.

# Training compute configurations

Open **Training**. Its tabs separate **Training jobs**, **Training configurations** and **Compute configurations**. A compute configuration specifies the machine; a training configuration specifies the hyperparameters. Keep these reusable settings separate from the job that executes them.

## Before you begin

Identify the dataset and the training workload. Use the CPU and memory specifications in the machine selector to compare the available training machines. Machine availability depends on your environment.

## Create a training compute configuration

1. Open **Model Lab > Training > Compute configurations** and select **Create**.

2. Enter a **Configuration name**.

3. Search for and select a **Machine type**. The selector displays machine names together with CPU and memory specifications.

4. Select **Create** after completing the required fields, then confirm that the saved compute configuration appears in the list.

Choose a machine for the dataset and training workload. Training jobs require a training compute configuration; processing compute used for ingestion or onboarding is not interchangeable with it. The machine sizes shown below illustrate the available choices.

![Training compute configuration with a searchable machine type selector.](/_fern-files/_img/48ef8d20d60de349b714ee22c8853c490e187f508bf9cd2ca2c38b8d0b8fd395.webp)

## Review and reuse a configuration

Open the saved record and check the configuration name, ID, compute type and machine type. Select this configuration in **Training jobs > Create**. The job's maximum runtime is set on the job, independently of the machine configuration.

Create a separate compute configuration when you want to compare another machine. Retain the configuration reference with each run so that runtime and model comparisons account for the compute used.

## Next steps

Set the learning behaviour in [Training configuration](/ui-guides/models/training-configuration), then create a [Training job](/ui-guides/models/training-jobs).