The leading global research institute, Institute of Technological Innovation, affiliated with the Advanced Technology Research Council in Abu Dhabi, has announced the launch of the second version of its large linguistic model named Falcon 2 today.
Falcon 2 features two advanced versions; Falcon 2 11B linguistic model, which operates more efficiently and is trained on 5.5 trillion token codes and 11 billion variable factors, and the second version of Falcon 2 11B (LEM), which boasts a “vision-to-language” feature that efficiently converts visual inputs into textual outputs.
Both models are equipped with multilingual operational capability.
The Falcon 2 11B (LEM) model is the first of its kind launched by the Institute of Technological Innovation as a multimodal model, the only model in the market capable of transforming visual images into written texts, marking a significant advancement in artificial intelligence innovations.
The performance of Falcon 2 11B was compared against several other artificial intelligence models in the same category, showing superior performance to the new Llama 3 model from Meta by a margin of 8 billion variable factors. It also achieved performance close to Google’s GPT-7. These results are derived from the Hugging Face list, a globally leading platform that employs objective evaluation tools and designs evaluation lists for open-source large linguistic models.
Both Falcon 2 11B and 11B (LEM) versions are open-source, aiming to empower developers worldwide to access and benefit from them limitlessly.
Expansion and development of the next generation of Falcon 2 models to offer a variety of sizes and scopes are planned soon.
These models will be enhanced with advanced machine learning techniques such as the Mixture of Experts (MoE) system, aiming to elevate performance to higher levels.
It is worth noting that all artificial intelligence models developed by the Institute of Technological Innovation have received high world rankings and are considered among the best open-source linguistic models.
The new and advanced Falcon 2 11B models are expected to better meet market needs, especially in the continuously evolving field of advanced generative artificial intelligence.
The Falcon 2 11B models have been optimized to operate in multiple languages, effortlessly interacting with tasks in languages such as English, French, Spanish, German, Portuguese, and others, enhancing their versatility and efficiency in various scenarios.
As for Falcon 2, it includes the “vision-to-language” system capable of identifying and analyzing visual images to offer a wide range of applications in various sectors. Whether in healthcare, finance, e-commerce, education, law, where it can be used for document management, digital archiving, supporting visually impaired individuals, and other useful applications.
These models can efficiently run on a single graphics processing unit, making them easy to develop, deploy, and integrate into small devices such as mobile phones and other handheld devices.
Mr. Faisal Al Bannai, Secretary-General of the Advanced Technology Research Council and Advisor for Strategic Research Affairs and Advanced Technology to the President of the United Arab Emirates, stated that Falcon 2 11B model is the first release in the Falcon 2 series. Despite the distinguished performance of Falcon 2 11B model, he expressed commitment to developing open-source models and achieving Falcon’s institution goals.
New multimodal models are expected to be developed in various sizes and introduced to the market soon, aiming to empower.Developers and organizations aim to enhance their privacy and access one of the best artificial intelligence models to improve the use of AI technology.
Dr. Hakim Hassid, the acting executive director and head of research at the Artificial Intelligence unit at the Institute of Technological Innovation, stated that in this rapidly evolving field, developers are aware of the great benefits they will achieve from flexible and small-sized language models, in addition to their role in reducing computer energy needs and complying with sustainability standards. He pointed out that these models are characterized by their flexibility and ease of use with advanced artificial intelligence infrastructure, which is the future destination for technological advancements.
He emphasized that the new capabilities of Falcon 2 in analyzing images and converting them into written text are a new opportunity to use artificial intelligence and enhance user experience in this promising field.
He explained that the diversity and flexibility of the Falcon 2 11B model will drive the Institute of Technological Innovation to expand to work on bigger and more exciting innovations in the field of artificial intelligence, such as adopting a new type of machine learning capability known as the “expert mix,” which involves integrating small networks with multiple specialties to ensure collaboration across different fields to provide advanced and customized responses, resembling a team of intelligent assistants who can predict or make decisions when needed.
This approach contributes to improving accuracy levels and speeding up decision-making processes, opening the door to designing more intelligent and efficient artificial intelligence systems.
It should be noted that the FALCON 2.11B model is licensed under the Institute of Technological Innovation’s Falcon 2.0 license, which is based on the Apache 2.0 software license containing an acceptable use policy that encourages responsible use of artificial intelligence.
