The Technology Innovation Institute (TII) has launched “Falcon LLM,” a 40 billion-parameter foundational large language model (LLM), advancing its generative AI credentials.
The AI intercentre unit of the AI and Digital Sciences Research Center (AIDRC), the team that built Noor, the world’s largest Arabic language model, built the Falcon LLM, a 40B model that outperforms GPT3.
The Falcon LLM 40B model is trained on one trillion tokens. The model uses only 75% of GPT-3, 40% of Chinchilla and 80% of PaLM-62B.
A wide range of
LLM can be used in a wide range of applications, such as chatbots, virtual assistants, language translation, content generation, and sentiment analysis. They can help businesses streamline customer service operations by providing efficient and effective responses to customer inquiries. This deep technology enables companies and countries to save huge amounts of money—from improving efficiency to reducing labor costs and identifying new revenue streams—simply by implementing them in a business or sector.
Dr. Ray O Johnson, CEO of TII, said: “2023 is proving to be the year of AI. Falcon LLM is a milestone announcement for us, but it’s just the beginning. By the end of the year, we’ll be sharing about News of a substantial increase in capabilities in this area.
“We know this is the beginning of an important journey,” Johnson continued, “and we will continue to provide the region with its own AI success stories, aligned with the UAE’s National AI Strategy.”
main impact
Professor Mérouane Debbah, Principal Investigator at the AI and Digital Sciences Research Centre, said: “We are excited to advance the world’s understanding of the power and benefits of these LLMs, which will have important implications in fields as diverse as education, healthcare, filmmaking, and CGI .This is an important milestone in the field as the country’s economy continues to grow and diversify. The Falcon LLM model is just the beginning of a new journey.”
Dr Ebtesam Almazroui, Director of AIDRC’s AI Inter-Centre Unit and LLM Program Leader, who played a key role in building LLM and enhancing the UAE’s capabilities in this field, said: “This achievement is a testament to the UAE’s forward-thinking approach and its Emphasis on innovation and technology.The development of this model is part of our national economic vision and strategic plan “We the UAE 2031”.
“Our Falcon LLM 40B model will significantly outperform models such as BLOOM and GPT-3, even at a fraction of the size. Thanks to state-of-the-art data pipelines, Falcon LLM also improves new generation models. For example, it is compatible with Chinchilla (from DeepMind) and PaLM-62B (from Google) are comparable in performance and much cheaper to train. The Falcon LLM is trained with the scaling law of Hoffmann et al., and the size of the model is kept relatively modest, the data Unprecedented quality.” — trade arab news agency