license: mit
MaLLM-Bench - Malaysian Exam Benchmark Dataset
Dataset Summary
The Malaysian Exam Benchmark Dataset is a collection of questions derived from Malaysia’s national examinations, designed to evaluate the performance of Large Language Models (LLMs) in understanding the Malay language and its cultural nuances. This dataset addresses gaps in existing benchmarks like MMLU and its translations by incorporating culturally relevant and contextually rich content.
The dataset includes questions from primary and secondary school exams such as Sijil Pelajaran Malaysia (SPM), Penilaian Menengah Rendah (PMR), Pentaksiran Tingkatan Tiga (PT3), Ujian Pencapaian Sekolah Rendah (UPSR). The subjects span various areas of study, such as history, geography, and Islamic study, making it a well-rounded resource for evaluating LLMs on tasks that require both language understanding and cultural relevance.
| Subject | Question Count |
|---|---|
| Accounting | 793 |
| Biology | 311 |
| Business Study | 731 |
| Chemistry | 432 |
| Design and Technology | 100 |
| Economic Study | 314 |
| General Study | 220 |
| Geography | 161 |
| History | 1287 |
| Islamic Study | 216 |
| Malay Language | 1363 |
| Physical Education | 144 |
| Physics | 311 |
| Science | 318 |
Dataset Structure
Data Instances
An example from general study looks as follows:
{
"question": "Antara yang berikut, yang manakah bidang kuasa bersama Kerajaan Persekutuan dan Kerajaan Negeri?",
"options": [
"Pendidikan",
"Tanah",
"Perumahan",
"Perpustakaan"
],
"answer": 3,
"subject": "general_study"
}