Italy strengthens its role in artificial intelligence with the launch of Minerva 7B, the new version of the language model developed by Sapienza University of Rome within the FAIR (Future Artificial Intelligence Research) framework, funded by the PNRR and developed in collaboration with CINECA, which provided access to the Leonardo supercomputer.
Minerva 7B is a fully Italian Large Language Model (LLM), created by the Sapienza NLP research group led by Prof. Roberto Navigli. With 7 billion parameters—more than double the previous version—it represents a significant advancement in natural language understanding and generation in Italian. The model has been trained on a massive dataset of 1.5 trillion words and refined through advanced instruction tuning, enabling it to produce more accurate, relevant, and safe responses while reducing issues such as hallucinations and ensuring appropriate outputs.
Innovation and collaboration
The project, developed within the PNRR framework, involved 15 researchers and PhD candidates from Sapienza, alongside key partners such as CINECA and NVIDIA. The official presentation featured contributions from Sapienza Rector Antonella Polimeni, FAIR President Giuseppe De Pietro, CINECA General Director Alessandra Poggiani, and Prof. Navigli, who demonstrated the model’s capabilities through a live demo.
Key statements:
- Antonella Polimeni: “Minerva confirms Sapienza’s commitment to combining academic rigor and innovation, serving society and the future.”
- Roberto Navigli: “Minerva is a starting point for developing transparent and secure solutions, with the goal of expanding collaborations with publishers and public institutions.”
- Alessandra Poggiani: “With CINECA, we support cutting-edge projects for a more inclusive digital society.”
- Giuseppe De Pietro: “Minerva is not only a highly valuable scientific achievement, but also a successful example of cooperation within the FAIR partnership. Supporting outstanding research outputs like Minerva is central to FAIR’s mission. We believe Minerva has the potential to become a reference Large Language Model for Public Administration, and we will continue to promote its development.”
An open and inclusive future
Minerva is already accessible to the public via minerva-llm.org and will be available for download in the coming weeks. Its open-source approach makes it particularly suitable for adoption by Public Administrations and other institutional sectors.
With the launch of Minerva 7B, Italy not only raises the bar for national language models, but also lays the foundation for a more inclusive and transparent technological development, strengthening the role of Italian research in the global AI landscape.
For more information, visit the official page.