Overview

arcee-ai/Meraj-Mini is an open-source, 7 billion parameter instruction-tuned language model, fine-tuned from Qwen2.5-7B-Instruct. It is specifically developed to enhance Arabic language capabilities while maintaining robust English proficiency. The model underwent a rigorous development process including data preparation with translated English datasets, iterative training, and evaluation of 15 different variants to achieve optimal bilingual performance.

Key Capabilities

• Arabic Language Understanding: Excels in general comprehension, reading comprehension, and common-sense reasoning tailored for Arabic.
• Cultural Adaptation: Generates content incorporating Arabic cultural nuances.
• Bilingual Proficiency: Demonstrates top-tier performance in Arabic benchmarks and competitive results in English.
• Mathematics and Coding: Supports mathematical reasoning and code generation in Arabic.

Benchmarks and Performance

Arcee Meraj Mini consistently outperforms state-of-the-art models on the Open Arabic LLM Leaderboard (OALL) and shows superior performance on Translated MMLU. Its English performance is comparable to leading models, indicating effective retention of English language knowledge while specializing in Arabic.

Good For

• Developing advanced Arabic-speaking chatbots and virtual assistants.
• Generating high-quality, culturally relevant Arabic content for various needs.
• Creating personalized educational experiences for Arabic speakers.
• Applications requiring mathematical reasoning and code generation in Arabic.