---
license: mit
task_categories:
- text-retrieval
- sentence-similarity
language:
- en
tags:
- rag
- chunking
source_datasets:
- The Pile
---

# FreeChunk Corpus

This dataset is derived from **The Pile** and is used for evaluating and training chunking models, specifically for the FreeChunk framework.

## Dataset Structure

The dataset consists of documents split into sentences.

### Features

- `uuid`: Unique identifier for the document.
- `sentences`: List of sentences in the document.
- `source`: Source of the document (e.g., from The Pile).
- `original_token_count`: Number of tokens in the original document.

## Usage

```python
from datasets import load_dataset

dataset = load_dataset("XiaSheng/FreeChunk-corpus")
```