metadata

license: mit
task_categories:
  - question-answering
language:
  - tr
tags:
  - mmlu
  - benchmark
  - multiple_choice
pretty_name: Turkish Grammer MMLU
size_categories:
  - 1K<n<10K

Turkish-Grammar-MMLU

This dataset, created by Turkish-DB, is a multiple-choice question-answering (QA) dataset covering Turkish grammar topics. It is designed to evaluate model performance on various Turkish grammar subjects, similar to the MMLU (Massive Multitask Language Understanding) benchmark.

Overview

Name: Turkish-Grammar-MMLU
Provider: Turkish-DB
Task: Multiple-Choice QA
Modality: Text
Format: CSV (also accessible via API in Parquet format)
Language: Turkish
Size: 3,700 examples
Splits: Single train split (3.7K samples)
Source: https://www.edebiyatogretmeni.org/
License: MIT

Data Structure

When downloaded, the dataset consists of a CSV file with the following columns:

Column	Description
`source`	URL or document name where the question was sourced
`classes`	Sub-grammar rule category (e.g., phonetics)
`subject`	General topic heading (28 distinct subjects)
`question`	Question text
`choices`	List of 5 answer options
`answer`	Index of the correct option (0–4)

Usage

Load the dataset using the datasets library:

from datasets import load_dataset

dataset = load_dataset("turkish-db/turkish-grammer-mmlu", split = "train")
print(dataset)

Expected output:

Dataset({
    features: ["source", "classes", "subject", "question", "choices", "answer"],
    num_rows: 3700
})

Example Query

# Inspect the first example
print(dataset[0])

Sample output:

{
    'source': 'https://www.edebiyatogretmeni.org/ses-bilgisi-test-1/',
    'classes': 'ses-bilgisi',
    'subject': 'ses-bilgisi',
    'question': 'Aşağıdaki cümlelerin hangisinde "ünsüz düşmesine" uğramış birden çok sözcük kullanılmıştır?',
    'choices': [
        'Sevginin küçücük görünümünden sıcacık bir hikâye çıkardığını görmüştüm.',
        'Ufacık dedikodular bile büyük bir aileyi dağıtmak için yeterli olur.',
        'Evlerin minicik saçaklarından yere kadar buzlar sarkıyordu.',
        'Araba, büyücek bir binanın önünde sarsılarak durdu.',
        'Mahallenin daracık sokaklarında çeşitli oyunlar oynardık.'
    ],
    'answer': 0
}

Citation

When using this dataset in your work, please cite:

@misc{turkish-grammar-mmlu,
  title   = {Turkish-Grammar-MMLU},
  author  = {{Turkish-DB}},
  year    = {2025},
  url     = {https://huggingface.co/datasets/turkish-db/turkish-grammer-mmlu}
}

License

This dataset is released under the MIT License.