News
G42 Company Inception Launches 20+ Arabic AI Language Models
The suite of tools is based on the 70-billion-parameter JAIS 70B large language model with bilingual Arabic and English AI capabilities.
Inception, a company that develops advanced AI models backed by G42, has launched JAIS 70B, a large language model (LLM) specially tailored for Arabic natural language processing (NLP).
The 70-billion parameter model was created to enhance the integration of generative AI across sectors, including content creation, customer service, and data analytics.
The new model offers open-source bilingual AI capabilities in Arabic and English. The technology was developed through rigorous training and fine-tuning of 370 billion tokens, including the largest Arabic dataset ever used in an open-source foundational model, with around 330 billion Arabic tokens.
Along with JAIS 70B, Inception has also announced the launch of 20 AI models ranging from 590 million to 70 billion parameters. The new tools are capable of addressing the needs of small-scale applications as well as complex enterprise solutions.
The AI models have been fine-tuned for chat applications using 1.6 trillion tokens in Arabic and English. Dr. Andrew Jackson, Inception’s CEO, highlighted the company’s passion for preserving Arabic language and heritage while democratizing AI access:
“Releasing JAIS 70B and this new family of models reinforces our commitment to delivering the highest-quality AI foundation model for Arabic-speaking nations,” he noted.
Also Read: Top Free AI Chatbots Available In The Middle East
After launching JAIS-13B and JAIS-13B-chat in August 2023, Inception introduced JAIS-30B and JAIS-30B-chat, with a focus on Arabic-centric models. JAIS 70B and JAIS 70B-chat have both demonstrated superior performance in English and Arabic benchmarks compared to the earlier models.
Neha Sengupta, Principal Applied Scientist at Inception, explained that JAIS variants were trained from scratch and outperformed others that were adapted from existing models. Inception also used Llama2 to enhance efficiency, reduce computational complexity and lessen environmental impact when developing JAIS 70B.