Global-MMLU-Lite

Global MMLU Lite (from Cohere For AI) is a multilingual variant of MMLU: a compact question set in many languages evaluated with human- and machine-verified translations. It shows broad knowledge in each language at a fraction of the compute cost.

Global-MMLU-Lite measures performance in languages other than english. In detail, it checks:

• multilingual knowledge breadth
• quality across many languages with verified translations
• cultural sensitivity of the question sets
• efficient comparisons across many languages

The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better.

What is the Global-MMLU-Lite benchmark?

Global MMLU Lite (from Cohere For AI) is a multilingual variant of MMLU: a compact question set in many languages evaluated with human- and machine-verified translations. It shows broad knowledge in each language at a fraction of the compute cost.

What does Global-MMLU-Lite test?

Global-MMLU-Lite mainly measures performance in languages other than English. Concretely, it checks: multilingual knowledge breadth; quality across many languages with verified translations; cultural sensitivity of the question sets.

How is the Global-MMLU-Lite score calculated?

The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better. The declared maximum score of this benchmark is 100.0.

Where do I get Global-MMLU-Lite data and scores?

The official source is https://huggingface.co/datasets/CohereForAI/Global-MMLU-Lite. The official benchmark page/reference is linked on this page. Every score on Open LLM Radar is also available as open data: download the scores JSON (http://open-llm-radar-api:8000/api/v1/benchmarks/global-mmlu-lite/scores) or pull it directly from the Open LLM Radar API with any HTTP client.