Friday, 10th of January, a tutorial on Large Language Models (LLMs) was conducted by the Language Technology Group (LTG) at UiO. Andrey Kutzov and David Samuel began the workshop with talks on the foundations of deep learning and transformer technology, explaning how these models are employed in pre-training to achieve of seemingly human-like conversational capabilities.
Egil Rønningstad then conducted a hands-on tutorial, assisted by David and Andrey, during which 50 virtual machines were set up on the Norwegian AI Cloud (NAIC), with open-source large language models for Norwegian (the NORA.LLM model family) using Python. This setup enabled possible for all the students to log on to NAIC and be up and running with the models in a couple of minutes.
The students developed a solid understanding of the fundamentals of modern generative language models and acquired practical experience of loading, using, and evaluating LLMs on NAIC, rather than relying on querying black-box API endpoints like ChatGPT.
Thus, this workshop showcases the advantages of the concerted efforts of the NORA collaboration, with NAIC providing computing resources and LTG sharing their expertise. It all came together at the well-organized winter school in Tromsø, hosted by SFI Visual intelligence / the Machine Learning Group as part of the NORA Norwegian AI Research School.
A big thank you to everyone involved, especially to Visual Intelligence, for yet another excellent NLDL conference and winter school. Thank you!