INESC TEC is the first Portuguese entity to join international consortium to accelerate the use of generative AI in science and engineering

Creating an open ecosystem that brings together representatives from academia, research centres, supercomputing centres and companies that are developing, training and using large-scale Artificial Intelligence (AI) models, as well as building and operating large-scale computing systems – this is the goal of TPC, The Trillion Parameter Consortium. INESC TEC was the first Portuguese entity to join this partnership, which gathers more than 70 entities and aims to promote the sharing of experiences, tools and data, collaboration between teams, and best practices in the responsible development of AI.

The development of large-scale AI models poses challenges for the scientific community; from the outset the high computational costs that training models with millions, billions and, eventually, trillions of parameters entails, becoming a serious constraint for research in this area. However, it also brings a series of opportunities. To address these challenges and fully explore said opportunities, a group of entities from different parts of the globe came together to accelerate the use of generative AI in science and engineering.

The Trillion Parameter Consortium (TCP) aims - through collaboration between experts in the fields of AI and high-performance computing - to build an open and diverse community that brings together entities that develop large-scale generative AI models and/or operate computing systems.

"The TPC brings together three major groups of entities. First, those who work in the development, training and evaluation of large-scale models. Then, entities that design and build hardware and software systems. Finally, the organisations that will use the AI systems stemming from this collaboration to address questions in science, engineering, healthcare and other domains", explained Rui Oliveira, researcher at INESC TEC - emphasising that the objective of this alliance is to unite efforts to boost the impact of many projects and initiatives already underway.

This way, the partnership will promote the sharing of experiences, tools and data, collaboration between academia, R&D centres, companies and supercomputing centre, while encouraging best practices in the development, integration and responsible evaluation of AI. Collaborative projects, meetings, hackathons, seminars, visits, white papers or summer schools are some of the many activities planned by the consortium.

"By integrating a global network of resources, knowledge and experiences, we are also helping to address a major challenge for the development of AI today, which is open access – including data, source codes or workflows. This access is fundamental to progress in all areas related to generative AI", said Rui Oliveira.

INESC TEC's participation in the TPC was formalised during the Consortium's kick-off meeting in Europe, which took place at the Barcelona Supercomputing Center in June - bringing together the European AI, high-performance computing and research communities. The presentation of several projects, like the "AuroraGPT: Building an Interdisciplinary Foundation Model" (operated by the Aurora supercomputer), took place during the event. INESC TEC presented the latest work related to "Growing and Training a diverse, global AI For Science Community", as well as international projects featuring the Institute's participation:  BigHPC , RISC2, EPICURE and HANAMI – towards developing high-performance computing.

This article was first published on 11 July by INESC TEC.