29-04-2024

Almawave has developed “Velvet”, an Italian open-source, multilingual, and multimodal AI model trained with Cineca supercomputing infrastructure. The AI model is characterized as:

• Multilingual: trained in the primary European languages, as well as Brazilian Portuguese and Swahili,
• With a specific focus on the Italian language and content.
• Multimodal: text, voice, and images
• Compliant with the AI Act, the Data Act, and privacy regulations

With Velvet it will be possible to develop vertical models for different sectors, in which Almawave has well-established expertise when it comes to data and the relevant domain.

The objective is to create a foundational Large Language Model (LLM) (meaning that it will serve as the technological basis for other models developed subsequently), which will primarily focus on the Italian language and Italian content.

Technological and applicative expertise come together in this initiative with Cineca’s unique know-how in supercomputing infrastructures, resulting in a partnership that will continue to grow over time. The Interuniversity Consortium, made up of 118 public bodies including 2 Ministries and 70 Italian Universities, manages Leonardo, one of the world’s most advanced high-performance scientific computing systems. Currently ranked sixth in the top 500 list of the world’s most powerful supercomputers and second in Europe, Leonardo is a key node of the European EuroHPC network.

Valeria Sandei, CEO of Almawave: “This partnership looks to the future and will lead us to the development of highly strategic AI technologies, whose impact and virtues could benefit many sectors that we’re already quite familiar with, such as public administration, transportation, healthcare, tourism, the green transition, and finance, in which we have already completed hundreds of artificial intelligence projects using our own technology. This is just the beginning of an exciting new phase in our evolution and we are proud to be facing it with Cineca.”

Francesco Ubertini, President of Cineca: “Velvet will have the unique characteristic of being multimodal, but also, and above all, multilingual, thus giving an international dimension to an AI model created and trained in Italy.”