Course provider: University of Ljubljana, Faculty of Computer and Information Science (UL FRI)
Instructors: Aleš Žagar (UL FRI), Matej Klemen (UL FRI), Domen Vreš (UL FRI), Tjaša Arčon (UL FRI), Živa Štebljaj (UL FRI), Marko Robnik-Šikonja (UL FRI)
Learning objectives: Get practical knowledge on how to use large language models (LLMs) via API calls, their fine-tuning for specific tasks, and efficient inference.
Course contents: To use large language models (LLMs) in a research or business setting, it is necessary to i) to access them on local computers or on GPU servers via API, ii) adapt them to specific needs by fine-tuning them. The hands-on workshop will present theoretically and practically, step-by-step:
- How LLMs can be used via API on a local or remote server.
- How to fine-tune encoder-only models such as BERT.
- How to fine-tune generative models such as Llama, Gemma, and GaMS using compute-efficient techniques such as LoRA.
- How to implement efficient LLM inference.
The work will be based on practical datasets and problems from business and research environments. The use cases will cover popular tasks such as sentiment prediction, information retrieval, and question answering.
Learning outcomes: Practical knowledge of LLM API use and fine-tuning. Practical problem-solving skills with LLMs.