---
dataset_info:
  features:
  - name: Unique_Serial
    dtype: string
  - name: subject
    dtype: string
  - name: question
    dtype: string
  - name: correct_answer
    dtype: string
  - name: options
    dtype: string
  - name: question_char_count
    dtype: int64
  - name: train_test_val
    dtype: string
  splits:
  - name: train
    num_bytes: 40552418
    num_examples: 111148
  - name: validation
    num_bytes: 5053686
    num_examples: 13916
  - name: test
    num_bytes: 5067426
    num_examples: 13885
  download_size: 19828710
  dataset_size: 50673530
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
license: mit
task_categories:
- question-answering
language:
- bn
size_categories:
- 100K<n<1M
---
# BnMMLU: Measuring Massive Multitask Language Understanding in Bengali

**BnMMLU** is a multiple-choice Bengali-language benchmark for Massive Multitask Language Understanding. It covers various academic and general-knowledge subjects (e.g., Accounting, Mathematics, Science, Humanities) with each example consisting of:

* (`string`): Unique question identifier.
* (`string`): Domain/subject of the question (e.g., Accounting).
* (`string`): The Bengali question text.
* (`list<string>`): Four answer choices in Bengali.
* (`string`): The correct option label (`"a"`, `"b"`, `"c"`, or `"d"`).
* (`int`): Character length of the question.
* (`string`): Split indicator (`"train"`, `"validation"`, or `"test"`).


If you are using the dataset, please cite it using:
```
@misc{joy2025bnmmlumeasuringmassivemultitask,
      title={BnMMLU: Measuring Massive Multitask Language Understanding in Bengali}, 
      author={Saman Sarker Joy},
      year={2025},
      eprint={2505.18951},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2505.18951}, 
}
```