> ## Documentation Index
> Fetch the complete documentation index at: https://arkor-92aeef0e-eng-615.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Dataset

> Tell the trainer where data comes from.

# Dataset

The `dataset` field on [`createTrainer`](/guides/sdk/create-trainer) takes one of two shapes: a HuggingFace repository name or a blob URL.

## HuggingFace (most projects)

```ts theme={null}
dataset: {
  type: "huggingface",
  name: "arkorlab/triage-demo",
}
```

Public Hub repos work without further auth. The bundled templates (`triage`, `translate`, `redaction`) all use this form. Optional `split` and `subset` let you target a specific split or named subset.

## Blob URL (your own data)

```ts theme={null}
dataset: {
  type: "blob",
  url: "https://example.com/data.jsonl",
  token: process.env.DATASET_TOKEN, // optional
}
```

Use this when the data lives somewhere you control: a signed S3 URL, an internal CDN, anything the backend can `GET` once at the start of the run.

Local files are not in `DatasetSource` today. To use one, host it as a blob URL or upload it to a private HuggingFace repo first.

## Reference

For the full discriminated union, every field, the `token` semantics, and the rationale behind picking each form, see the [`DatasetSource` reference](/sdk/dataset).