Loading datasets...
Release Date: 7/23/2026
Format: CSV
Size: 2.04 MB
Share
MMLU-ProX Sindhi is a translated version of the MMLU-ProX benchmark, adapted into Sindhi for evaluating large language models on multilingual reasoning and knowledge understanding. It preserves the original multiple choice structure across 11,739 questions spanning STEM, Social Sciences, Humanities, and Other domains, with detailed subdomain coverage including mathematics, physics, law, economics, psychology, business, biology, and more. The dataset enables systematic benchmarking of LLM capabilities in Sindhi, a low resource South Asian language.
Restrictions/Special Constraints
This dataset is a Sindhi translation and derivative of the MMLU-ProX benchmark. It is distributed under the MIT License, consistent with the licensing of the original MMLU-ProX dataset. If you redistribute this dataset or modified versions, you should retain the original copyright notice and MIT license, provide attribution to the original MMLU-ProX authors, and clearly indicate that this release is a translated derivative.
Forbidden Usage
There are no additional restrictions beyond those specified by the MIT License. Users should not misrepresent benchmark results obtained on this translated dataset as directly comparable to the original English MMLU-ProX benchmark without clearly disclosing that evaluation was performed on a translated version. Proper attribution to the original MMLU-ProX authors and this Sindhi translation should be maintained in redistributed versions.
Each example contains: id, question (Sindhi), options (A–D), correct_key, domain, subdomain, source.
Domain distribution:
STEM: 6,557
Social Sciences: 3,385
Humanities: 878
Other: 919
Top subdomains: Mathematics (1,350), Physics (1,298), Chemistry (1,132), Engineering (968), Law (958), Economics (841), Psychology (797), Business (789), Biology (716)
Derived from li-lab/MMLU-ProX, translated into Sindhi.
Machine-translated; may contain translation artifacts or ambiguities.
Some source concepts/terminology may not have exact Sindhi equivalents.
Results may not be directly comparable to the original English MMLU-ProX benchmark.
Released under the MIT License.
This dataset is a Sindhi translation and derivative of the MMLU-ProX benchmark. The original MMLU-ProX dataset is distributed under the MIT License, and this translated version retains the same licensing terms.
Please cite both the original MMLU-ProX benchmark and this Sindhi translation when using the dataset in research, evaluation, or model development.
