horelulus/ID_REG_QA_Small

---
license: apache-2.0
---

# 🧾 Indonesian Legal QA Dataset

This repository contains a **question-answer (QA) dataset** generated from parsed Indonesian regulations, focusing on **legal quoting and comprehension**. Designed to facilitate legal-aware LLMs, the dataset provides direct QA mappings to individual articles for contextual understanding and reference.

---

## 📌 Dataset Highlights

* **Source**: Generated from the [ID\_REG\_Parsed](https://huggingface.co/datasets/Azzindani/ID_REG_Parsed) repository
* **Format**: QA pairs based on individual articles (no chunking)
* **Scale**: Augmented by applying 10 QA templates across suitable regulation entries
* **Filtering**: Programmatic filtering removes redundant or overly broad article explanations
* **Target Use**: Train/test LLMs for **regulation comprehension**, **legal quoting**, and **document-level QA**

---

## ⚙️ Pipeline Overview

* **Environment**: Executed in a single Jupyter Notebook on **Kaggle Cloud**

* **Data Flow**:

1. **Pull** parsed articles from `ID_REG_Parsed`
2. Filter and refine results for clarity and legal context
3. Apply **template-driven QA generation** (10 variations)
4. **Push** QA dataset directly to this repository

* **Performance**:

* Completed in \~20 minutes using Kaggle GPU resources
* Cloud-to-cloud transfer without local storage dependency

---

## 🧠 Use Cases

* Fine-tuning LLMs for **legal question answering**
* Benchmarks for **article referencing and quoting**
* Few-shot prompting for legal search assistants
* Legal text evaluation with grounded answers

---

## ⚠️ Disclaimer

This dataset is intended for **research and development** only. QA pairs are generated synthetically from publicly available legal text and may not reflect official interpretations.

---

## 🙏 Acknowledgments

* **[Hugging Face](https://huggingface.co/)** for hosting open datasets
* **[Kaggle](https://www.kaggle.com/)** for compute and cloud-to-cloud capabilities

---