Abu Dhabi, 5th August, 2020 (WAM) — Inception, a G42 company specializing in the development of advanced artificial intelligence (AI) models and applications, today launched its latest large language model, GIS 70B, along with 20 other AI models to support Arabic-based natural language processing.
The company said in a statement today that the “JIS 70B” model, which contains 70 billion parameters, was designed for developers of Arabic-based natural language processing solutions, and promises to accelerate the integration of generative AI services across various sectors, enhancing capabilities in areas such as customer service, content creation, and data analysis.
The company also unveiled a comprehensive suite of core and fine-tuned GIS models; comprising 20 models, across 8 sizes, ranging from 590 million to 70 billion parameters, fine-tuned specifically for chat applications, trained on up to 1.6 trillion language units in Arabic, English, and programming data.
This set of GIS models accommodates a wide range of uses, and aims to accelerate innovation, development and research opportunities for many future applications for the Arabic-speaking and bilingual Arab community.
“Today, AI has become a value-adding force, and large language models have been at the forefront of the rise in AI adoption,” said Dr. Andrew Jackson, CEO of Inception. “JIS was created to preserve the Arabic heritage, culture and language, and to make AI accessible to everyone.”
“The release of GIS 70B and this new family of models reinforces our commitment to providing a high-quality AI baseline model for Arab countries,” he added, noting that the training and adaptation techniques we successfully deliver for Arabic language models are scalable to other underserved languages.
“For models up to 30 billion parameters, we were able to train JESS from scratch and consistently outperformed community-produced models,” said Neha Sengupta, principal applied scientist at Inception. “For models with 70 billion parameters and above, the computational complexity and environmental impact of training them from scratch was significant.”
- For more: Elrisala website and for social follow us on Facebook
- Source of information and images “wam“
