NightPrince/mofaser-quran-tafsir

---
language:
- ar
- en
license: cc-by-4.0
task_categories:
- question-answering
- text-generation
tags:
- quran
- tafsir
- islam
- arabic
- instruction-tuning
- fine-tuning
pretty_name: Mofaser — Quran Tafsir QA Dataset
size_categories:
- 100K<n<1M
---

# Mofaser — Quran Tafsir QA Dataset

**243,129 records** covering all **6,236 Quranic verses** interpreted through **18 classical tafsir sources** (8 Arabic + 10 English), formatted as a rich QA dataset for fine-tuning and instruction-tuning LLMs.

## Dataset Summary

Each record pairs a Quranic verse with a tafsir explanation in a QA format with varied question phrasings. The dataset includes full metadata: surah/ayah identifiers, Arabic and English verse text, Meccan/Medinan classification, juz number, author name, and a pre-formatted ChatML conversation column.

## Splits

| Split | Records |
|---|---|
| train | 219,096 |
| val | 11,991 |
| test | 12,042 |

Splits are grouped by ayah — all templates/tafsirs for a given verse stay in the same split, preventing data leakage.

## Columns

| Column | Description |
|---|---|
| `id` | Unique record identifier |
| `language` | `ar` or `en` |
| `surah_number` | Surah index (1–114) |
| `surah_name_ar` | Arabic surah name |
| `surah_name_en` | English surah name |
| `surah_type` | Meccan or Medinan |
| `juz_number` | Juz index (1–30) |
| `ayah_number` | Verse number |
| `ayah_text` | Arabic Quranic text (Tanzil) |
| `ayah_text_en` | English translation (Saheeh International) |
| `tafsir_slug` | Tafsir source identifier |
| `tafsir_name` | Tafsir human-readable name |
| `tafsir_author` | Scholar/author name |
| `system_prompt` | Varied system prompt |
| `question` | Question (varied phrasing, 18 AR + 15 EN templates) |
| `answer` | Tafsir explanation text |
| `answer_length` | Character count of answer |
| `conversation` | Pre-formatted ChatML turn |
| `question_template_id` | Template index used |
| `source` | Data attribution |

## Arabic Tafsirs
Ibn Kathir · Al-Saddi · Al-Baghawi · Tanwir Al-Miqbas · Al-Wasit · Al-Tabari · Muyassar · Al-Qurtubi

## English Tafsirs
Ibn Kathir (abridged) · Tazkirul Quran · Kashf Al-Asrar · Al-Qushairi · Kashani · Al-Tustari · Asbab Al-Nuzul · Ibn Abbas · Al-Jalalayn · Maarif-ul-Quran

## Usage

```python
from datasets import load_dataset

ds = load_dataset("NightPrince/mofaser-quran-tafsir")

# Arabic records only
ar = ds["train"].filter(lambda x: x["language"] == "ar")

# Load the ChatML conversation column directly for training
print(ds["train"][0]["conversation"])
```

## Data Sources
- Quran text: [Tanzil Project](https://tanzil.net)
- English translation: Saheeh International via [risan/quran-json](https://github.com/risan/quran-json)
- Tafsir: [spa5k/tafsir_api](https://github.com/spa5k/tafsir_api)

## Citation
If you use this dataset, please cite the original tafsir sources and the Tanzil Project.