Fine-tuned Small Language Models (SMLs) - faster, cheaper, private(er)

Margareth is exploring how Small Language Models (SMLs) could be faster, cheaper, and more private alternatives to Large Language Models (LLMs), without sacrificing performance where it counts. DistilBERT and Mistral-7B are better suited for real-world deployment compared to giant models, especially when handling narrow, specialized tasks. However, using small models effectively requires careful task matching and proper fine-tuning, since they are not general-purpose geniuses.

Margareth is a master's student in Business Information Technology from Estonia, currently completing an innovation and management internship at Lyse Tele (mentored by Celina Thom). As part of her master’s thesis, she is exploring the fine-tuning of small and large language models to evaluate the potential of AI usage in crisis communication.

(Spots are limited and people end up on a waitlist, so register early!)

We meet at Itera office in Strandgaten 18, 5013 Bergen on the 4th floor. Doors should be opened, if not search for Dalibor Jovic in Itera on the intercom.
We will order some pizza, so please make sure you register on time, as we will close registration before the event. Also, please cancel the attendance if situation changes.