The Technology Innovation Institute (TII) in Abu Dhabi has taken a significant leap forward in artificial intelligence by launching its second-generation large language model (LLM) series, Falcon 2. This new series boasts two models:

Falcon 2 11B: This efficient and accessible LLM is trained on a massive dataset of 5.5 trillion tokens and boasts 11 billion parameters. It’s multilingual, tackling tasks in English and several other languages.

Falcon 2 11B VLM is TII’s first multimodal model. It can understand text and interpret visual inputs, converting images to text. 

This opens doors for applications in healthcare, finance, and more.

Independent testing confirms Falcon 2’s prowess: It surpasses Meta’s recently launched Llama 3 and performs on par with Google’s leading model, Gemma 7B. Both Falcon 2 versions are open-source and freely available for developers worldwide.

Beyond performance, TII prioritizes accessibility. Falcon 2 models can run efficiently on a single graphics processing unit (GPU), making them ideal for integration into even personal devices like laptops. This scalability opens doors for broader adoption.

Committed to open-source development, TII plans to expand the Falcon 2 family with various sizes and functionalities. 

These future models will incorporate advanced techniques like a “Mixture of Experts.” This technique essentially creates a team of specialized AI sub-models that work together, boosting accuracy and efficiency in decision-making.

Falcon 2’s potential applications are vast, spanning document management, e-commerce, education, and accessibility tools for the visually impaired.

H.E. Faisal Al Bannai, Secretary General of ATRC, emphasized their commitment to open-source AI: “Our aim is to ensure that developers and entities that value their privacy have access to one of the best AI models to enable their AI journey