Based in the United Arab Emirates (UAE), Technological Innovation Institute (TII) announced a new iteration of its Large Language Model (LLM).
New Falcon 2 includes a template with image-to-text conversion capability, TII said in a Monday, May 13. Press release.
The Falcon 2 11B VLM, which is one of two versions of the new LLM, has Vision Language Model (VLM) capabilities that allow it to convert visual inputs into textual outputs, according to the release.
Potential applications of VLM capabilities include document management, digital archiving, contextual indexing and assisting the visually impaired, the release said.
The other new version, Falcon 2 11B, is designed to be a “more efficient and accessible LLM,” the statement said. It was trained on 5.5 trillion tokens with 11 billion parameters and matches or exceeds the performance of AI models in its class among pre-trained models.
Both models are multilingual, allowing them to handle tasks in English, French, Spanish, German, Portuguese, and various other languages, depending on the version. Both are also open source, giving developers around the world unlimited access.
Because they can run on a single graphics processing unit (GPU), both can be integrated into laptops and other devices, according to the release.
“AI is continually evolving and developers are recognizing the countless benefits of smaller, more efficient models,” said Dr. Hakim Hacid, Executive Director and Acting Chief Investigator of TII’s AI Cross-Center Unit, said in the release. “In addition to reducing computing power requirements and meeting sustainability criteria, these models provide increased flexibility, integrating seamlessly with cutting-edge AI infrastructure, the next emerging megatrend.”
There is a tendency to develop smaller and more cost-effective systems. AI models that make AI more accessible to businesses.
“Smaller LLMs give users more control compared to large language models like ChatGPT or Anthropic’s Claude, making them more desirable in many cases. » Brian Petersonco-founder and CTO of Numeric keypad, a cloud-based, AI-powered platform, told PYMNTS in an interview published in March. “They are able to filter a smaller subset of data, making it faster, more affordable and, if you have your own data, much more customizable and even more accurate. »
