---
pretty_name: BnMMLU
language:
- bn
tags:
- bengali
- benchmark
- evaluation
- multiple-choice
- multilingual
size_categories:
- 100K<n<1M
configs:
- config_name: full
  data_files:
  - split: train
    path: data/full/bnmmlu-full.parquet
- config_name: hard
  data_files:
  - split: train
    path: data/hard/bnmmlu-hard.parquet
---

# BnMMLU

BnMMLU is a Bengali multiple-choice benchmark for measuring multitask language understanding across 41 subjects.

This repository exposes two dataset configs:

- `full`: the complete benchmark with 134,382 questions.
- `hard`: a 15,074-question subset containing the most challenging items, including a `failure_ratio` field.

The source data in this workspace is stored as CSV. For the Hugging Face dataset repo, the upload script converts the files to Parquet so that `options` is preserved as a real list instead of a stringified Python list.

## Dataset Schema

### `full`

| Column | Type | Description |
| --- | --- | --- |
| `Unique_Serial` | `int64` | Unique sequential identifier. |
| `subject_name` | `string` | Subject or domain label. |
| `question` | `string` | Bengali question text. |
| `correct_answer` | `string` | Correct option key: `a`, `b`, `c`, or `d`. |
| `options` | `list[string]` | Four answer options in order. |
| `question_char_count` | `int64` | Character count of the question text. |

### `hard`

The `hard` config contains all columns from `full` plus:

| Column | Type | Description |
| --- | --- | --- |
| `failure_ratio` | `float64` | Proportion of evaluated models that answered incorrectly. |

## Loading

```python
from datasets import load_dataset

full = load_dataset("samanjoy2/BnMMLU", "full")
hard = load_dataset("samanjoy2/BnMMLU", "hard")
```

## Citation

```bibtex
@inproceedings{joy-shatabda-2026-bnmmlu,
    title = "{B}n{MMLU}: Measuring Massive Multitask Language Understanding in {B}engali",
    author = "Joy, Saman Sarker  and
      Shatabda, Swakkhar",
    editor = "Liakata, Maria  and
      Moreira, Viviane P.  and
      Zhang, Jiajun  and
      Jurgens, David",
    booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {ACL} 2026",
    month = jul,
    year = "2026",
    address = "San Diego, California, United States",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2026.findings-acl.593/",
    doi = "10.18653/v1/2026.findings-acl.593",
    pages = "12211--12230",
    ISBN = "979-8-89176-395-1",
    abstract = "Large-scale multitask benchmarks have driven rapid progress in language modeling, yet most emphasize high-resource languages such as English, leaving Bengali underrepresented. We present BnMMLU, a comprehensive benchmark for measuring massive multitask language understanding in Bengali. BnMMLU spans 41 domains across STEM, humanities, social sciences, and general knowledge, and contains 134,375 multiple-choice question{--}option pairs-the most extensive Bengali evaluation suite to date. The dataset preserves mathematical content via MathML, and includes BnMMLU-HARD, a compact subset constructed from questions most frequently missed by top systems to stress difficult cases. We benchmark 24 model variants across 11 LLM families, spanning open-weights general/multilingual, Bengali-centric open-weights, and proprietary models, covering multiple parameter scales and instruction-tuned settings. We evaluate models under standardized protocols covering two prompting styles (Direct vs. Chain-of-Thought) and two context regimes (0-shot vs. 5-shot), reporting accuracy consistently across families. Our analysis highlights persistent gaps in reasoning and application skills and indicates sublinear returns to scale across model sizes. We release the dataset and evaluation templates to support rigorous, reproducible assessment of Bengali language understanding and to catalyze progress in multilingual NLP."
}
```
## Dataset Status

> **Warning:** This dataset currently contains a preliminary, non-final version of BnMMLU. The dataset may undergo further corrections, filtering, restructuring or other modifications before the final release. Users should verify the data carefully and use this version at their own risk. Results produced using the current version may not be directly comparable with results obtained from future releases.