ai4bharat/Indic-Rag-Suite

🌏 Multilingual Indic RAG Suite

A comprehensive multilingual question-answering dataset covering 18 Indian languages with 21,439,886 total samples, designed for RAG (Retrieval-Augmented Generation) applications and multilingual NLP research.

🚀 Quick Start

from datasets import load_dataset

# Load specific language (recommended)
dataset = load_dataset("ai4bharat/Indic-Rag-Suite", "as")
train_data = dataset['train']

print(f"Loaded {len(train_data)} samples")

# Access examples
for example in train_data[:2]:
    print(f"Q: {example['question']}")
    print(f"A: {example['answer']}")
    print("---")

📊 Dataset Overview

Available Languages (18 total)

Code	Language	Load Command
as	Assamese	load_dataset('ai4bharat/Indic-Rag-Suite', 'as')
bn	Bengali	load_dataset('ai4bharat/Indic-Rag-Suite', 'bn')
en	English	load_dataset('ai4bharat/Indic-Rag-Suite', 'en')
gu	Gujarati	load_dataset('ai4bharat/Indic-Rag-Suite', 'gu')
hi	Hindi	load_dataset('ai4bharat/Indic-Rag-Suite', 'hi')
kn	Kannada	load_dataset('ai4bharat/Indic-Rag-Suite', 'kn')
ks	Kashmiri	load_dataset('ai4bharat/Indic-Rag-Suite', 'ks')
mai	Maithili	load_dataset('ai4bharat/Indic-Rag-Suite', 'mai')
ml	Malayalam	load_dataset('ai4bharat/Indic-Rag-Suite', 'ml')
mni	Manipuri	load_dataset('ai4bharat/Indic-Rag-Suite', 'mni')
mr	Marathi	load_dataset('ai4bharat/Indic-Rag-Suite', 'mr')
ne	Nepali	load_dataset('ai4bharat/Indic-Rag-Suite', 'ne')
or	Odia	load_dataset('ai4bharat/Indic-Rag-Suite', 'or')
pa	Punjabi	load_dataset('ai4bharat/Indic-Rag-Suite', 'pa')
sat	Santali	load_dataset('ai4bharat/Indic-Rag-Suite', 'sat')
ta	Tamil	load_dataset('ai4bharat/Indic-Rag-Suite', 'ta')
te	Telugu	load_dataset('ai4bharat/Indic-Rag-Suite', 'te')
ur	Urdu	load_dataset('ai4bharat/Indic-Rag-Suite', 'ur')

Dataset Statistics

• Total Samples: 21,439,886
• Languages: 18
• Source: Wikipedia + Llama 3.3 70B
• Task: Question Answering with Reasoning

📋 Data Structure

Each sample contains:

• question: The question text
• answer: The answer text
• reasoning: Detailed reasoning/explanation
• paragraph: Source Wikipedia paragraph
• title: Wikipedia article title
• wiki_id: Wikipedia article ID
• url: Wikipedia URL
• language: Language code
• text: Combined question+answer+reasoning for RAG

💡 Usage Examples

Load Single Language

from datasets import load_dataset

# Most efficient - load only what you need
hindi_dataset = load_dataset("ai4bharat/Indic-Rag-Suite", "hi")
train_data = hindi_dataset['train']

# Convert to pandas for analysis
import pandas as pd
df = train_data.to_pandas()
print(f"Dataset shape: {df.shape}")

Load Multiple Languages

languages = ['hi', 'bn', 'ta', 'en']
datasets = {}

for lang in languages:
    datasets[lang] = load_dataset("ai4bharat/Indic-Rag-Suite", lang)
    print(f"{lang}: {len(datasets[lang]['train'])}")

---

Dataset URL: https://huggingface.co/datasets/ai4bharat/Indic-Rag-Suite
Total Samples: 21,439,886
Languages: 18
Last Updated: 2025-06-07