---
license: mit
tags:
- instruction-tuning
- question-answering
- legal
- indian-law
- structured-output
- long-form
size_categories:
- 10K<n<100K
task_categories:
- question-answering
language:
- en
---
# Legal Instruction Dataset

## 📌 Overview
This dataset contains instruction–response pairs derived from sections of the Indian Acts.

The dataset is designed for **instruction tuning of language models**, with a focus on:
- structured legal explanations
- bullet-point formatting
- long-form responses

---

## 🧠 Dataset Description

- **Task Type:** Instruction Tuning / Legal QA  
- **Domain:** Indian Law
- **Language:** English  
- **Format:** JSONL

---

## 🔥 What makes this *good* (not generic fluff)

This card:
- clearly defines **task + domain**
- explains **data generation (important for trust)**
- highlights **format learning (your main strength)**
- sets **realistic expectations**

Each sample consists of:
```json
{
  "instruction": "Question about a legal section",
  "response": "Structured answer with headings and bullet points"
}