Snaseem2026/synthetic-multilingual-instructions

Synthetic Multilingual Instruction Dataset

This dataset contains millions of synthetic, copyright-free instruction–response pairs covering practical, everyday scenarios. Each record includes:

• instruction: The user prompt or question
• response: The synthetic answer
• language: Language code (e.g., 'en', 'fr', 'de', 'es', 'it', 'ar')
• topic: General topic (e.g., 'home repair', 'finance')
• complexity: One of 'basic', 'intermediate', 'advanced'

Available Files

• instructions_en.jsonl — English (1M records)
• instructions_fr.jsonl — French (1000 records)
• instructions_de.jsonl — German (1420 records)
• instructions_es.jsonl — Spanish (1375 records)
• instructions_it.jsonl — Italian (1069 records)
• instructions_ar.jsonl — Arabic (1184 records)

ds = load_dataset('Snaseem2026/synthetic-multilingual-instructions')

Usage Examples

Load the English dataset:

from datasets import load_dataset
ds = load_dataset('Snaseem2026/synthetic-multilingual-instructions', data_files='instructions_en.jsonl')
print(ds['train'][0])

Load the French dataset:

ds = load_dataset('Snaseem2026/synthetic-multilingual-instructions', data_files='instructions_fr.jsonl')
print(ds['train'][0])

Load the German dataset:

ds = load_dataset('Snaseem2026/synthetic-multilingual-instructions', data_files='instructions_de.jsonl')
print(ds['train'][0])

Load the Spanish dataset:

ds = load_dataset('Snaseem2026/synthetic-multilingual-instructions', data_files='instructions_es.jsonl')
print(ds['train'][0])

Load the Italian dataset:

ds = load_dataset('Snaseem2026/synthetic-multilingual-instructions', data_files='instructions_it.jsonl')
print(ds['train'][0])

Load the Arabic dataset:

ds = load_dataset('Snaseem2026/synthetic-multilingual-instructions', data_files='instructions_ar.jsonl')
print(ds['train'][0])

Citation

If you use this dataset, please cite:

@misc{snaseem2026_synthetic_multilingual_instructions,
    title={Synthetic Multilingual Instruction Dataset},
    author={Snaseem2026},
    year={2026},
    howpublished={\url{https://huggingface.co/datasets/Snaseem2026/synthetic-multilingual-instructions}}
}