The Arabic AI model Jais is officially open source, with 13 billion parameters
The UAE research team recently announced the open source Arabic large-scale model Jais .
Jais is a 13 billion parameter pre-trained Arabic and English bilingual large-scale language model , trained on a dataset containing 72 billion Arabic tokens and 279 billion English/code tokens. The model was jointly developed by Cerebras, the UAE University of Artificial Intelligence and Inception, a subsidiary of G42.
Jais is named after the highest peak in the United Arab Emirates. Timothy Baldwin, a professor at the United Arab Emirates University of Artificial Intelligence, said that since there is not enough Arabic data to train a Jais-sized model, the computer code in the English data can help train the reasoning ability of the model.
The model is now open source and users can get it from HuggingFace.