Arabic Leaderboards: Introducing Arabic Instruction Following, Updating AraGen, and More

Hugging Face Blog · 1 year ago

Inception launched Arabic-Leaderboards, a unified platform developed with Mohammed bin Zayed University of Artificial Intelligence to consolidate evaluations of Arabic language AI models. The AraGen-03-25 benchmark expanded from 279 to 340 question-answer pairs, with a new Arabic Instruction Following leaderboard powered by the Arabic IFEval Benchmark as the first publicly available instruction-following evaluation for Arabic. The platform enables researchers to assess Arabic LLM performance across generative tasks and instruction adherence, with plans to expand to additional modalities and invite community contributions.