Base project for development of next generation large scale multimodal language models

Purpose and goal: ** Denna text är maskinöversatt **

Based on previous experience, the project will develop next-generation multimodal large-scale language models. Through the previous project with Swedish language models, we know there is a need for more powerful multimodal models in both the private and public sectors. There are good opportunities for Sweden to take a leading role.

We have built up both competence and collaborations that enable continued development, and we have also identified users as well as applications that will benefit from powerful multimodal models. Expected results and effects: ** Denna text är maskinöversatt **

Result from this project is the first major multimodal language model for the Nordic languages. All models will be open and available for research and commercialization. Through this, we see great potential for increased use of AI within Swedish organizations and that the model enables the creation of an active start-up ecosystem that builds generative AI applications.

Approach and implementation: ** Denna text är maskinöversatt **

The project will be implemented in three phases, with control and decision points between each phase. The control points are based on the KPIs defined for each phase, as well as the development in the outside world (both technical and regulatory). Each phase contains a number of different work packages (APs), which in some cases span several different phases.

AP1 - framework AP2 - data AP 3 - training framework and model architecture AP4 - components