The “Inception,” the AI center in the “G42” group, announced the launch of the open-source version of the “JIES” model, a high-quality Arabic language model that is considered the first of its kind in the world.
“JIES” has been built on 13 billion indicators and has been trained on a new dataset containing 395 billion tokens in both Arabic and English languages.
The “JIES” model, named after the highest peak in the United Arab Emirates, will utilize state-of-the-art generative AI capabilities in the Arab world.
This model is the result of collaboration between the “Inception” center, Mohammed bin Zayed University for Artificial Intelligence, the world’s first specialized university for AI research, and “Cyberbas Systems” company.
The model was trained using the “Condor Galaxy,” a supercomputer powered by AI capable of computations at high rates, reaching up to a million trillion calculations per second. This computer was designed in collaboration between “G42” and “Cyberbas Systems.”
The launch of “JIES” is a crucial milestone in the field of artificial intelligence in the Arab world. This model, developed in the UAE capital Abu Dhabi, offers more than 400 million Arabic speakers a significant opportunity to explore the potential of generative AI, as well as enhancing the UAE’s position as a leading center in AI, innovation, cultural preservation, and international cooperation.
By releasing the source of the “JIES” model, the “Inception” center aims to enhance the involvement of scientific communities, academia, and developers in promoting the growth of a vibrant AI ecosystem in Arabic and boosting innovation in this field. “JIES” may become a benchmark model for other languages currently underrepresented in advanced AI environments.
Andrew Jackson, CEO of the “Inception” AI center, stated that collaboration is the foundation for success. They are setting new standards for AI development in the Middle East, emphasizing the importance of the Arabic language and its rich heritage in the field of AI. The “JIES” initiative reflects their commitment to excellence, innovation, and widespread dissemination of AI systems.
Professor Eric Zing, President of Mohammed bin Zayed University for Artificial Intelligence, stated that developing a large linguistic model for the Arabic language requires advanced AI research and a deep understanding of the language’s diversity and rich heritage. He emphasized the importance of large linguistic models in various societal aspects that are increasingly important. He also affirmed that Mohammed bin Zayed University for Artificial Intelligence will continue to lead in effective and accurate large linguistic models.
“JIES” is a significant linguistic model based on transformers that harnesses various advanced features, including the Attention Bias feature “ALBAY” that enables the model to infer long sequences for providing better and more accurate context. The model also uses leading-edge technologies like the activated Swiggle linear boundary units and setting maximum boundary parameters for updating to enhance the efficiency of model training and accuracy.
A team from Mohammed bin Zayed University for Artificial Intelligence and the “Inception” center evaluated and modified the open-source “JIES” model. The model was trained on a dedicated dataset containing 116 billion Arabic tokens to address the complexities, variations, and richness of the language. The recent development of the high-powered AI computer “Condor Galaxy 1” (CG-1) in collaboration between “G42” and “Cyberbas Systems” was utilized in this process. Furthermore, the dataset included 279 billion English tokens to enhance the model’s performance in the. . .
Switching between languages. The “Inspection” center and the Mohammed bin Zayed University for Artificial Intelligence will continue to enhance and expand the “GES” model to meet the growing needs of the user community.In his statement, Andrew Fieldman, CEO of Cyberpass Systems, said: “Our strategic partnership with “G42” has already yielded excellent results, as we introduced the powerful “Condor Galaxy” computer powered by artificial intelligence with ultra-fast computing capabilities a few weeks ago. Today, this partnership presents another exceptional achievement, which is the large-scale Arabic language model aimed at the open-source software community. At Cyberpass, we are committed to developing advanced technologies and exploring innovative use cases. “GES” is a significant contribution to the global open-source software community, serving as concrete evidence of the ease of use of the “Condor Galaxy 1″ computer and its rapid capability to develop artificial intelligence models.”