This repository provides tools for translating the Massive Multitask Language Understanding (MMLU) dataset from English to Norwegian, along with evaluation scripts.
- Research Protocol: Details the translation process, quality scoring, and evaluation strategy.
- Translation Quality Evaluation: Evaluation of translation quality.
- Translation Scripts: Tools to accurately translate MMLU questions while preserving structure and meaning.
- Evaluation Tools: Built upon lm-evaluation-harness to assess translation quality and model performance on both Norwegian and English datasets.
The MMLU dataset includes over 14,000 multiple-choice questions across 57 subjects. High-quality translations ensure that the original difficulty and context are maintained for Norwegian audiences.
This project is licensed under the MIT License.