GPT-SW3: a foundational model for Swedish NLP
| Reference number | |
| Coordinator | Lindholmen Science Park AB - AI Sweden |
| Funding from Vinnova | SEK 6 147 484 |
| Project duration | September 2022 - September 2024 |
| Status | Completed |
| Venture | AI - Leading and innovation |
| Call | Advanced and innovative AI |
Important results from the project
The purpose of the project was to evaluate GPT-SW3, a family of large-scale Swedish language models, to assess its applicability in various tasks within the public sector and industry. The goal of testing and validating the models in practical use cases has been achieved through collaborations with multiple partners who have tested the model in areas such as summarization, categorization, and text generation. The results have provided valuable insights into the potential and limitations.
Expected long term effects
The project has resulted in a deeper understanding of how GPT-SW3 and the underlying technology can be applied in practice. Several partners have identified potential use cases but also challenges such as limited performance compared to other models, legal obstacles, and technical constraints. The project has thus contributed to knowledge building around large-scale language models in Sweden and created conditions for future development and implementation of AI technology across various sectors.
Approach and implementation
The project was executed through collaborations with the project partners, who tested GPT-SW3 within their respective organizations. This setup allowed for a broad evaluation of the model in various contexts. However, some partners highlighted the need for better coordination and continuous communication to maximize knowledge exchange and efficiency. The project successfully identified key insights and established a network for future collaboration on AI and language models in Sweden.