quranlab/quran-tafsir

QuranLab — Multilingual Quran Tafsir Dataset

A ready-to-use collection of Quran commentaries and annotated translations, aligned to the canonical 6,236 ayahs.

QuranLab is a volunteer effort. Our aim is to present these works carefully and at high quality, and to help them travel faithfully — in the spirit in which they were written — not to claim them as ours.

The text here reaches you through the work of QuranEnc.com, Tafsir Center for Quranic Studies, Quranic Universal Library (QUL), spa5k/tafsir_api and OpenITI RELEASE. Who to thank, and what we consulted without ever quoting, is set out in SOURCES.md.

Highlights

• 109 configs
• 33 languages
• 679,724 rows
• One consistent row for every ayah in every config
• Focused on Ahl al-Sunnah tafsir works

Choose the work and language you need from the config selector.

Popular configs

Config	Language	Content
tafsir-mukhtasar-tr	Turkish	Al-Mukhtasar
tafsir-saadi-tr	Turkish	Tafsir al-Sa'di
tafsir-mukhtasar-en	English	Al-Mukhtasar
tafsir-rowwad-annotated-en	English	Rowwad translation with notes
tafsir-noor-annotated-en	English	Noor International translation with notes
tafsir-hilali-khan-annotated-en	English	Hilali-Khan translation with notes
tafsir-ibnkathir-ar	Arabic	Tafsir Ibn Kathir
tafsir-tabari-ar	Arabic	Tafsir al-Tabari

The full config list is available in the Hub selector and in metadata/tafsirs.csv.

Usage

from datasets import load_dataset

dataset = load_dataset(
    "quranlab/quran-tafsir",
    "tafsir-mukhtasar-tr",
    split="train",
)

print(dataset[0]["verse_key"])
print(dataset[0]["text"])

Switch the config name to load another tafsir or language.

Main columns

• verse_key: ayah key such as 2:255
• text: tafsir text, or translation with notes for annotated configs
• lang, tafsir_id: language and config identity
• has_commentary, coverage_status: whether this ayah actually has commentary
• structural fields: surah, ayah, juz, page, surah_name_ar/en, …

The work's author, title, school, source and terms are the same for every ayah, so they live once per config in metadata/tafsirs.csv instead of on all 6,236 rows.

Every config uses the same ayah order, from 1:1 to 114:6. It can be joined with quranlab/quran using verse_key.

Notes

Annotated-translation configs are clearly named with annotated. If a source does not provide commentary for an ayah, the row remains present and coverage_status explains it.

Different configs have different reuse terms; they are listed per config in metadata/tafsirs.csv and LICENSES.md.

Questions and corrections are welcome in the dataset's Community tab.

For agents and pipelines

Deterministic facts an automated consumer needs, so nothing has to be inferred from a sample.

• Config naming: tafsir-{work}-{lang}
• Join key: verse_key — identical order in every config, and the same key used by quranlab/quran
• Default config: tafsir-ibnkathir-ar
• Format: Parquet, one directory per config, split train.
• Terms and credit are per config in metadata/, never per row — join on the config name when you need them.
• coverage_status is the honesty flag. A row whose source has no commentary for that ayah is present with empty text and a status explaining why — never back-filled from a neighbouring ayah.
• Configs ending in annotated are translations carrying explanatory notes, not standalone commentaries.

Inspect it without downloading the data:

# every config and split
curl https://datasets-server.huggingface.co/splits?dataset=quranlab/quran-tafsir

# a first page of rows
curl "https://datasets-server.huggingface.co/first-rows?dataset=quranlab/quran-tafsir&config=tafsir-ibnkathir-ar&split=train"

# machine-readable schema and provenance (Croissant JSON-LD)
curl https://huggingface.co/api/datasets/quranlab/quran-tafsir/croissant

Query it in place with SQL — no full download:

hf datasets sql quranlab/quran-tafsir "SELECT * FROM 'tafsir-ibnkathir-ar' LIMIT 5"

import duckdb
duckdb.sql("SELECT count(*) FROM 'hf://datasets/quranlab/quran-tafsir/tafsir-ibnkathir-ar/*.parquet'")

Quality, and how to check it yourself

Nothing is released until the integrity gate passes. Every claim below is something you can re-run rather than take on trust — here on the published data, not in our build tree:

What is guaranteed	Evidence
Every config is ayah-complete	6,236 rows per config in canonical order, 109 configs — no config is published partial
Source text is not paraphrased	each config declares its exact transformation in text_fidelity; most are verbatim, and every deviation is a documented, deterministic rendering
Sources are pinned, not loosely scraped	every source witness is fixed to an exact revision and checksum-verified before each build, so a silent upstream edit cannot slip in
Gaps are declared, never invented	where a source has no commentary the row is published empty and labelled source-unavailable; the build fails if those gaps ever change silently
Theology is classified per work	school and Ahl-al-Sunnah status recorded per config, with cautions (Isra'iliyyat, weak narrations) carried in the card rather than hidden
Release gate	1,781 automated checks — including live re-verification against the original publishers — must pass before anything is published

from datasets import load_dataset

ds = load_dataset("quranlab/quran-tafsir", "tafsir-ibnkathir-ar", split="train")

assert len(ds) == 6236
assert ds[0]["verse_key"] == "1:1" and ds[-1]["verse_key"] == "114:6"
# gaps are labelled, not silently empty
assert all(r or s for r, s in zip(ds["text"], ds["coverage_status"]))

Where a source is genuinely missing, the row is published empty and labelled rather than filled by guessing from a neighbouring entry. Coverage numbers in this card count only real content.

Citation

@misc{quranlab_quran_tafsir_2026,
  title        = {QuranLab — Multilingual Quran Tafsir Dataset},
  author       = {QuranLab},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/quranlab/quran-tafsir}}
}