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

# max_steps_per_epoch

#### `max_steps_per_epoch: <int>` (Optional)

## Description

`max_steps_per_epoch` controls the maximum number of steps performed in a single epoch during training.

* A step is defined as training your model on a single mini-batch of size `batch_size`
* An epoch is defined as the process of completing the specified number of steps, i.e., `max_steps_per_epoch` steps, after which validation is performed.

This is particularly useful for managing the trade-off between training time and validation frequency.

For example, to double the validation frequency while keeping the total number of mini-batches the model processes unchanged, halve `max_steps_per_epoch` and double `max_epochs`.

* `max_steps_per_epoch` must be `> 0`.

### Supported Task Types

* All

### Default Values

| run\_mode | Default Value |
| --------- | ------------- |
| FAST      | `1000`        |
| NORMAL    | `2000`        |
| BEST      | `2000`        |