# belebele by AI at Meta: AI Training Dataset
Source: https://savrn.com/datasets/belebele
Markdown alternate of the page above; the site index is https://savrn.com/llms.txt

---

## Dataset Card

By AI at Meta, published under cc-by-sa-4.0, revision 7899cdfa4e1e.

### The Belebele Benchmark for Massively Multilingual NLU Evaluation

Belebele is a multiple-choice machine reading comprehension (MRC) dataset spanning 122 language variants. This dataset enables the evaluation of mono- and multi-lingual models in high-, medium-, and low-resource languages. Each question has four multiple-choice answers and is linked to a short passage from the [FLORES-200](https://github.com/facebookresearch/flores/tree/main/flores200) dataset. The human annotation procedure was carefully curated to create questions that discriminate between different levels of generalizable language comprehension and is reinforced by extensive quality checks. While all questions directly relate to the passage, the English dataset on its own proves difficult enough to challenge state-of-the-art language models. Being fully parallel, this dataset enables direct comparison of model performance across all languages. Belebele opens up new avenues for evaluating and analyzing the multilingual abilities of language models and NLP systems.

[Read the full dataset card (1,750 words)](https://savrn.com/datasets/belebele/card)

## Structure

### acm_Arab 900 rows

| Split | Rows | Size |
| --- | --- | --- |
| test | 900 | 1.0 MB |

linkstringquestion_numberint64flores_passagestringquestionstringmc_answer1stringmc_answer2stringmc_answer3stringmc_answer4stringcorrect_answer_numstringdialectstringdstimestamp[s]

### afr_Latn 900 rows

| Split | Rows | Size |
| --- | --- | --- |
| test | 900 | 746.0 KB |

linkstringquestion_numberint64flores_passagestringquestionstringmc_answer1stringmc_answer2stringmc_answer3stringmc_answer4stringcorrect_answer_numstringdialectstringdstimestamp[s]

### als_Latn 900 rows

| Split | Rows | Size |
| --- | --- | --- |
| test | 900 | 817.8 KB |

linkstringquestion_numberint64flores_passagestringquestionstringmc_answer1stringmc_answer2stringmc_answer3stringmc_answer4stringcorrect_answer_numstringdialectstringdstimestamp[s]

### amh_Ethi 900 rows

| Split | Rows | Size |
| --- | --- | --- |
| test | 900 | 1.1 MB |

linkstringquestion_numberint64flores_passagestringquestionstringmc_answer1stringmc_answer2stringmc_answer3stringmc_answer4stringcorrect_answer_numstringdialectstringdstimestamp[s]

### apc_Arab 900 rows

| Split | Rows | Size |
| --- | --- | --- |
| test | 900 | 990.8 KB |

linkstringquestion_numberint64flores_passagestringquestionstringmc_answer1stringmc_answer2stringmc_answer3stringmc_answer4stringcorrect_answer_numstringdialectstringdstimestamp[s]

### arb_Arab 900 rows

| Split | Rows | Size |
| --- | --- | --- |
| test | 900 | 1.0 MB |

linkstringquestion_numberint64flores_passagestringquestionstringmc_answer1stringmc_answer2stringmc_answer3stringmc_answer4stringcorrect_answer_numstringdialectstringdstimestamp[s]

### arb_Latn 900 rows

| Split | Rows | Size |
| --- | --- | --- |
| test | 900 | 769.4 KB |

linkstringquestion_numberint64flores_passagestringquestionstringmc_answer1stringmc_answer2stringmc_answer3stringmc_answer4stringcorrect_answer_numstringdialectstringdstimestamp[s]

### ars_Arab 900 rows

| Split | Rows | Size |
| --- | --- | --- |
| test | 900 | 1.0 MB |

linkstringquestion_numberint64flores_passagestringquestionstringmc_answer1stringmc_answer2stringmc_answer3stringmc_answer4stringcorrect_answer_numstringdialectstringdstimestamp[s]

### ary_Arab 900 rows

| Split | Rows | Size |
| --- | --- | --- |
| test | 900 | 1.0 MB |

linkstringquestion_numberint64flores_passagestringquestionstringmc_answer1stringmc_answer2stringmc_answer3stringmc_answer4stringcorrect_answer_numstringdialectstringdstimestamp[s]

### arz_Arab 900 rows

| Split | Rows | Size |
| --- | --- | --- |
| test | 900 | 1.0 MB |

linkstringquestion_numberint64flores_passagestringquestionstringmc_answer1stringmc_answer2stringmc_answer3stringmc_answer4stringcorrect_answer_numstringdialectstringdstimestamp[s]

### asm_Beng 900 rows

| Split | Rows | Size |
| --- | --- | --- |
| test | 900 | 1.6 MB |

linkstringquestion_numberint64flores_passagestringquestionstringmc_answer1stringmc_answer2stringmc_answer3stringmc_answer4stringcorrect_answer_numstringdialectstringdstimestamp[s]

### azj_Latn 900 rows

| Split | Rows | Size |
| --- | --- | --- |
| test | 900 | 851.8 KB |

linkstringquestion_numberint64flores_passagestringquestionstringmc_answer1stringmc_answer2stringmc_answer3stringmc_answer4stringcorrect_answer_numstringdialectstringdstimestamp[s]

## Details

Repository

facebook/belebele

Publisher

AI at Meta

Task category

Question answering

Tags

Not stated by the source

Size category

100K<n<1M

Languages

af, am, ar, az, as, bm, bn, bo

Revision

7899cdfa4e1e0d733fd77c848e2c273cb1d32be2

Last updated

2024-08-12

## Files

127 files, 251.7 MB in total.

Data123 files · 251.6 MB

Documentation1 file · 25.5 KB

Other2 files · 4.9 KB

Repository1 file · 2.3 KB

Every file

| File | Type | Size | SHA-256 |
| --- | --- | --- | --- |
| data.zip | Data | 27.1 MB | 3e66552376c7 |
| data/acm_Arab.jsonl | Data | 2.8 MB | — |
| data/afr_Latn.jsonl | Data | 885.0 KB | — |
| data/als_Latn.jsonl | Data | 1.1 MB | — |
| data/amh_Ethi.jsonl | Data | 2.2 MB | — |
| data/apc_Arab.jsonl | Data | 2.7 MB | — |
| data/arb_Arab.jsonl | Data | 2.8 MB | — |
| data/arb_Latn.jsonl | Data | 895.2 KB | — |
| data/ars_Arab.jsonl | Data | 2.8 MB | — |
| data/ary_Arab.jsonl | Data | 2.8 MB | — |
| data/arz_Arab.jsonl | Data | 2.8 MB | — |
| data/asm_Beng.jsonl | Data | 3.2 MB | — |
| data/azj_Latn.jsonl | Data | 1.4 MB | — |
| data/bam_Latn.jsonl | Data | 1.1 MB | — |
| data/ben_Beng.jsonl | Data | 3.2 MB | — |
| data/ben_Latn.jsonl | Data | 884.8 KB | — |
| data/bod_Tibt.jsonl | Data | 4.2 MB | — |
| data/bul_Cyrl.jsonl | Data | 3.4 MB | — |
| data/cat_Latn.jsonl | Data | 959.4 KB | — |
| data/ceb_Latn.jsonl | Data | 940.9 KB | — |
| data/ces_Latn.jsonl | Data | 1.1 MB | — |
| data/ckb_Arab.jsonl | Data | 3.2 MB | — |
| data/dan_Latn.jsonl | Data | 907.5 KB | — |
| data/deu_Latn.jsonl | Data | 965.5 KB | — |
| data/ell_Grek.jsonl | Data | 3.9 MB | — |
| data/eng_Latn.jsonl | Data | 836.1 KB | — |
| data/est_Latn.jsonl | Data | 898.4 KB | — |
| data/eus_Latn.jsonl | Data | 870.0 KB | — |
| data/fin_Latn.jsonl | Data | 985.0 KB | — |
| data/fra_Latn.jsonl | Data | 1.1 MB | — |
| data/fuv_Latn.jsonl | Data | 828.6 KB | — |
| data/gaz_Latn.jsonl | Data | 963.4 KB | — |
| data/grn_Latn.jsonl | Data | 1.0 MB | — |
| data/guj_Gujr.jsonl | Data | 3.1 MB | — |
| data/hat_Latn.jsonl | Data | 863.8 KB | — |
| data/hau_Latn.jsonl | Data | 908.0 KB | — |
| data/heb_Hebr.jsonl | Data | 2.5 MB | — |
| data/hin_Deva.jsonl | Data | 3.1 MB | — |
| data/hin_Latn.jsonl | Data | 895.5 KB | — |
| data/hrv_Latn.jsonl | Data | 896.8 KB | — |
| data/hun_Latn.jsonl | Data | 1.2 MB | — |
| data/hye_Armn.jsonl | Data | 3.6 MB | — |
| data/ibo_Latn.jsonl | Data | 1.1 MB | — |
| data/ilo_Latn.jsonl | Data | 955.3 KB | — |
| data/ind_Latn.jsonl | Data | 880.4 KB | — |
| data/isl_Latn.jsonl | Data | 1.1 MB | — |
| data/ita_Latn.jsonl | Data | 953.7 KB | — |
| data/jav_Latn.jsonl | Data | 854.9 KB | — |
| data/jpn_Jpan.jsonl | Data | 1.7 MB | — |
| data/kac_Latn.jsonl | Data | 1.0 MB | — |
| data/kan_Knda.jsonl | Data | 3.5 MB | — |
| data/kat_Geor.jsonl | Data | 3.6 MB | — |
| data/kaz_Cyrl.jsonl | Data | 3.4 MB | — |
| data/kea_Latn.jsonl | Data | 890.2 KB | — |
| data/khk_Cyrl.jsonl | Data | 3.4 MB | — |
| data/khm_Khmr.jsonl | Data | 4.0 MB | — |
| data/kin_Latn.jsonl | Data | 932.1 KB | — |
| data/kir_Cyrl.jsonl | Data | 3.4 MB | — |
| data/kor_Hang.jsonl | Data | 1.5 MB | — |
| data/lao_Laoo.jsonl | Data | 3.3 MB | — |
| data/lin_Latn.jsonl | Data | 889.7 KB | — |
| data/lit_Latn.jsonl | Data | 1.0 MB | — |
| data/lug_Latn.jsonl | Data | 874.1 KB | — |
| data/luo_Latn.jsonl | Data | 856.7 KB | — |
| data/lvs_Latn.jsonl | Data | 1.1 MB | — |
| data/mal_Mlym.jsonl | Data | 3.8 MB | — |
| data/mar_Deva.jsonl | Data | 3.3 MB | — |
| data/mkd_Cyrl.jsonl | Data | 3.4 MB | — |
| data/mlt_Latn.jsonl | Data | 1.0 MB | — |
| data/mri_Latn.jsonl | Data | 1.0 MB | — |
| data/mya_Mymr.jsonl | Data | 4.3 MB | — |
| data/nld_Latn.jsonl | Data | 894.3 KB | — |
| data/nob_Latn.jsonl | Data | 892.1 KB | — |
| data/npi_Deva.jsonl | Data | 3.2 MB | — |
| data/npi_Latn.jsonl | Data | 869.4 KB | — |
| data/nso_Latn.jsonl | Data | 1.0 MB | — |
| data/nya_Latn.jsonl | Data | 923.8 KB | — |
| data/ory_Orya.jsonl | Data | 3.4 MB | — |
| data/pan_Guru.jsonl | Data | 3.2 MB | — |
| data/pbt_Arab.jsonl | Data | 2.9 MB | — |
| data/pes_Arab.jsonl | Data | 3.0 MB | — |
| data/plt_Latn.jsonl | Data | 1.0 MB | — |
| data/pol_Latn.jsonl | Data | 1.0 MB | — |
| data/por_Latn.jsonl | Data | 971.1 KB | — |
| data/ron_Latn.jsonl | Data | 1.1 MB | — |
| data/rus_Cyrl.jsonl | Data | 3.6 MB | — |
| data/shn_Mymr.jsonl | Data | 4.5 MB | — |
| data/sin_Latn.jsonl | Data | 982.2 KB | — |
| data/sin_Sinh.jsonl | Data | 3.3 MB | — |
| data/slk_Latn.jsonl | Data | 1.1 MB | — |
| data/slv_Latn.jsonl | Data | 892.9 KB | — |
| data/sna_Latn.jsonl | Data | 905.4 KB | — |
| data/snd_Arab.jsonl | Data | 2.8 MB | — |
| data/som_Latn.jsonl | Data | 933.2 KB | — |
| data/sot_Latn.jsonl | Data | 946.1 KB | — |
| data/spa_Latn.jsonl | Data | 1.0 MB | — |
| data/srp_Cyrl.jsonl | Data | 3.2 MB | — |
| data/ssw_Latn.jsonl | Data | 904.8 KB | — |
| data/sun_Latn.jsonl | Data | 889.8 KB | — |
| data/swe_Latn.jsonl | Data | 947.7 KB | — |
| data/swh_Latn.jsonl | Data | 863.1 KB | — |
| data/tam_Taml.jsonl | Data | 3.9 MB | — |
| data/tel_Telu.jsonl | Data | 3.3 MB | — |
| data/tgk_Cyrl.jsonl | Data | 3.5 MB | — |
| data/tgl_Latn.jsonl | Data | 971.7 KB | — |
| data/tha_Thai.jsonl | Data | 3.4 MB | — |
| data/tir_Ethi.jsonl | Data | 2.2 MB | — |
| data/tsn_Latn.jsonl | Data | 975.0 KB | — |
| data/tso_Latn.jsonl | Data | 961.5 KB | — |
| data/tur_Latn.jsonl | Data | 1.1 MB | — |
| data/ukr_Cyrl.jsonl | Data | 3.4 MB | — |
| data/urd_Arab.jsonl | Data | 3.1 MB | — |
| data/urd_Latn.jsonl | Data | 965.7 KB | — |
| data/uzn_Latn.jsonl | Data | 905.0 KB | — |
| data/vie_Latn.jsonl | Data | 1.5 MB | — |
| data/war_Latn.jsonl | Data | 971.5 KB | — |
| data/wol_Latn.jsonl | Data | 918.2 KB | — |
| data/xho_Latn.jsonl | Data | 873.5 KB | — |
| data/yor_Latn.jsonl | Data | 1.4 MB | — |
| data/zho_Hans.jsonl | Data | 1.3 MB | — |
| data/zho_Hant.jsonl | Data | 1.3 MB | — |
| data/zsm_Latn.jsonl | Data | 889.6 KB | — |
| data/zul_Latn.jsonl | Data | 899.4 KB | — |
| README.md | Documentation | 25.5 KB | — |
| data/README | Other | 1.4 KB | — |
| unused/belebele.py | Other | 3.4 KB | — |
| .gitattributes | Repository | 2.3 KB | — |

## License and Download

License

cc-by-sa-4.0

Access

No access gate

[Download from AI at Meta](https://huggingface.co/datasets/facebook/belebele)

Released by AI at Meta through its official repository on Hugging Face. [Read the license](https://creativecommons.org/licenses/by-sa/4.0/).

## AI at Meta

[All models and datasets](https://savrn.com/model-publishers/facebook)

## Explore More

- [The dataset directory](https://savrn.com/datasets)
- [The model directory](https://savrn.com/models)
- [Research papers](https://savrn.com/papers)

## Source

- Repository metadata, read 2026-10-19.
- [AI at Meta website](https://ai.meta.com/resources/models-and-libraries/)
- [Hugging Face record](https://huggingface.co/datasets/facebook/belebele)
- [How the hub is built](https://savrn.com/model-hub/methodology)
