README.md

---
license: other
task_categories:
- text-generation
size_categories:
- 10K<n<100K
language:
- en
tags:
- chat
- conversation
- instruction-tuning
- fanfiction
- ao3
pretty_name: AO3 Chat
---

# ao3_chat

## Overview

A reformatted version of the dataset
[**`midwestern-simulation-active/ao3_random_subset`**](https://huggingface.co/datasets/midwestern-simulation-active/ao3_random_subset), converted into
chat-style and instruction-tuning examples suitable for training
chat-oriented language models.

Each example consists of:
- a **system** prompt derived from AO3 metadata
- a **user** prompt requesting the contents of a specific work
- an **assistant** response containing the full text (or a chunk thereof)

This dataset is intended for supervised fine-tuning (SFT) of chat models.

---

## Structure

The repository contains two representations of the same data:

### `jsonl/`
Line-delimited JSON files

### `parquet/`
Columnar Parquet files containing the same fields, suitable for
efficient loading with PyArrow, Pandas, or Hugging Face Datasets.

## Prompt Format

### System
Metadata extracted from AO3 entries (rating, fandoms, relationships,
characters, tags, etc.).

### User Prompt
A request to provide the full text of a specific fanfiction work by title.

### Assistant Response
The original fanfiction text, optionally split into multiple parts for
long works.

## Source

Original data:

[`midwestern-simulation-active/ao3_random_subset`](https://huggingface.co/datasets/midwestern-simulation-active/ao3_random_subset)

This dataset preserves the original text content and metadata, while
changing only the formatting.

## Notes

* This dataset contains copyrighted material.

* It is provided for research and experimentation purposes.

* Users are responsible for complying with AO3’s terms of service and
applicable copyright laws.

## Intended Use

- Chat model fine-tuning
- Long-form text generation research