Abu Dhabi’s G42 Unveils Groundbreaking Open Source Arabic Language AI Model – Jais
Dubai: Can an Arabic language learning model (LLM) be harnessed to fuel generative AI applications? The answer is yes. Introducing Jais, a novel open-source Arabic LLM, created right here in Abu Dhabi.
Jais, touted as the world’s most advanced open source Arabic LLM, has been officially introduced on Wednesday. This innovative project was spearheaded by Inception, a subsidiary of Abu Dhabi’s G42, in partnership with the Mohammed bin Zayed University of Artificial Intelligence and Silicon Valley’s Cerebras Systems.
How was Jais Developed?
The formidable Jais model, boasting an impressive 13 billion parameters, was meticulously crafted using a specialized dataset. This dataset encompassed a staggering 116 billion Arabic tokens to capture the intricate nuances of the Arabic language, coupled with 279 billion English word tokens to elevate its cross-lingual performance.
Dr. Andrew Jackson, the CEO of Inception, elucidated that Jais was trained on the groundbreaking Condor Galaxy AI supercomputer, a multi-exaFLOP marvel jointly engineered by G42 and Cerebras.
Named after the highest peak in the UAE, Jais was collaboratively developed by a team of academics and engineers who recognized the scarcity of large bilingual language models in Arabic. Dr. Jackson acknowledged the challenge posed by the limited availability of Arabic data online, estimating that only 1% of Arabic content can be found on the web.
Jais Outperforms its Peers
Dr. Jackson affirmed that Jais surpasses existing Arabic models such as Falcon from Abu Dhabi’s Technology Innovation Institute, Llama 2 from Meta Platforms, and Bloom by a significant margin. He highlighted its competitive stance even with English models of comparable size, despite being trained on considerably less English data.
“We are committed to introducing Jais across various sectors including government, finance, energy/climate, and healthcare,” stated Dr. Jackson.
Prominent entities such as the UAE Ministry of Foreign Affairs, Ministry of Industry and Advanced Technology, Department of Health – Abu Dhabi, ADNOC, Etihad Airways, First Abu Dhabi Bank, e&, and Mubadala Investment Company are collaborating with Inception to integrate Jais into their operations.
Accessibility and Exploration
Jais can be downloaded from Hugging Face, according to the developers. Additionally, users interested in exploring Jais can do so online by registering on the Jais website and receiving an invitation to access the interactive playground environment.
The Launch of Jais: A Milestone for Arabic AI
Jais’ introduction marks a significant stride forward for AI within the Arabic-speaking world. Originating in Abu Dhabi, this pioneering creation empowers Arabic speakers worldwide with cutting-edge generative AI capabilities.