Next generation infrastructure (NextGen)

By collaborating broadly within AI Sweden, experiences from building large-scale language models was combined with capabilities from other research areas within AI. The Natural Language Understanding (NLU) team, which developed Sweden's first large-scale language model, GPT-SW3, and work on european projects such as Open Euro LLM, wehre among the contributors, together with AI Sweden's experts in edge learning infrastructure. Industrial partners provided specialized hardware and software expertise.

The new infrastructure focused on a decentralized architecture to overcome the limitations of centralized systems, utilizing data centers from partner organizations managed by a central orchestrator based at AI Sweden. By leveraging existing expertise and resources, the project aimed to develop a future-proof prototype with significant performance improvements.

Expected benefits included better model performance, enhanced collaboration, increased privacy and security, cost efficiency, and hardware-agnostic solutions, which were anticipated to promote accessibility for academia, the public sector, and industry. 

The high commercialization potential included offerings such as infrastructure as a service, data solutions, and technology licensing. 

Overall, the project was expected to have a positive impact on the global foundation model ecosystem, driving innovation and productivity across multiple sectors.

For more information, contact