The State of Qatar has officially launched Fanar, a groundbreaking bilingual large language model (LLM), on the opening day of the World Summit AI in Doha. Named after the Arabic word for “lighthouse,” Fanar is designed to bridge the gap between global English models and Arabic-centric AI. Developed by the Qatar Computing Research Institute (QCRI) at Hamad Bin Khalifa University (HBKU) and sponsored by the Ministry of Communications and Information Technology (MCIT), Fanar delivers unparalleled linguistic precision, contextual comprehension, and cultural depth.
Fanar is powered by two advanced models: Fanar Star 7B, developed from scratch, and Fanar Prime, an 8.87-parameter model built on Google’s Gemma 2.9B. It incorporates retrieval-augmented generation (RAG) modules for real-time information retrieval, especially for Islamic and cultural queries.
Key Features of Fanar LLM
Fanar launches with six cutting-edge AI applications:
- Fanar Chat
A bilingual chatbot capable of handling multiple Arabic dialects, accepting text and voice prompts, and producing multimodal outputs such as text, speech, and images. - Taleem
A teaching assistant designed to help educators create lesson materials, including summaries, multimedia, and quizzes. - Akhbar AI
An AI-powered newsroom assistant that assists editors in content creation, including writing articles, interview plans, and visuals. - Allama
A government services chatbot leveraging RAG to answer complex questions about processes and services. - Talk to Your Book
An innovative app that lets users engage with virtual agents representing books for an interactive reading experience. - News Insights
A news analysis tool developed by Hyperthink Technology to provide actionable insights and analytics.
Unparalleled Training and Cultural Integration
The Fanar LLM was trained on 1.3 trillion tokens, with 40% Arabic, 50% English, and 10% code. A curated corpus of over 400 billion Arabic tokens includes data on Qatar’s heritage, traditions, and colloquial Arabic dialects, ensuring culturally nuanced and contextually accurate responses.
Collaborators in this effort include Qatar University, Qatar National Library, Al Jazeera, and the Ministry of Endowment and Islamic Affairs. The model even integrates specialized RAG modules for Islamic data queries.
Benchmark testing reveals Fanar’s competitive edge, outperforming models like Allam 7B and Jais 13B while closely rivaling the newer Jais 6p7b model launched in September.
Empowering Arabic-Speaking Communities
Fanar offers services such as:
- Arabic-English and English-Arabic translation
- Summarization and creative writing tools
- Virtual assistants tailored for public and private sectors
This innovation aims to empower institutions and businesses to engage more effectively with Arabic-speaking audiences while respecting cultural nuances.
Commitment to Innovation and User Feedback
QCRI has incorporated a user-feedback system to refine the model continually. Open access to Fanar models and APIs will soon be available, enabling broader adoption and innovation.
A Billion-Dollar Technology Vision
Fanar’s launch reflects Qatar’s commitment to advancing AI and technology innovation. In 2024 alone, Qatar has pledged $14.5 billion for R&D across sectors, including AI, life sciences, and energy transition. Major investments include:
- €10 billion in French technology sectors.
- £1 billion in UK climate and AI tech, including a joint Qatar-UK AI research initiative.
- $2.5 billion to promote domestic AI and innovation programs.
Fanar marks a pivotal step in establishing Qatar as a leader in bilingual generative AI, aligning with its vision to blend innovation with cultural authenticity.
—————————————————————————————————
About Innovation Streets:
Welcome to InnovationStreets.com! Get the latest insights from the finance and technology industries. Discover industry updates on fintech innovations, regulatory developments, industry insights, events, and collaborations between banks, insurers, and fintechs. A platform for C-level executives to submit thought leadership content for their peers. Subscribe for updates!
