> See the full index of documentation pages at https://docs.zettabyte.com/llms.txt.

# Datasets

> Create datasets to use as training and validation sets for fine-tuning.

Source: https://docs.zettabyte.com/docs/models-data-storage/datasets

Last updated: 2026-07-21

Datasets provide the training and validation sets used to fine-tune a model. The fine-tuning wizard's **Dataset Configuration** step needs both a **Training Set** and a **Validation Set**, so at least one dataset must exist first.

The **Datasets** tab lists your datasets with a **Show Only My** toggle and a search box. Each row shows these columns: Name/ID, Size, Status, Upload Progress, Latest Version, Updated Time, and User.

## Create a dataset

Select **Create Dataset** and set:

- **Dataset Name** (required): starts with a lowercase letter, and uses only lowercase letters, numbers, and hyphens.
- **Description**: optional.
- **Version**: set to `v1` at creation.
- **Storage Type**: **Shared Storage** or **S3 Storage** (see Storage).
- **Datasets Address** (required): the location of the data, with a **Refresh** control.
- **Version Description**: optional.

## Upload data first (Shared Storage)

For Shared Storage datasets, the data must already be in the shared volume (`PVC`) before you create the dataset. Upload your data into the shared volume first, then **Datasets Address** selects from that location and **Refresh** re-reads it. A `finetune_` prefixed zip uploaded into the shared volume is auto-extracted and becomes available as a fine-tuning dataset. An empty picker before any data is uploaded is expected, not an error.