Pimp My Llama

Prompt Engineering, RAG, Fine-Tuning - there’s many ways to customize the performance of LLMs for your use case. But which one is right for YOU?
GPT-4, Llama3 and many other Large Language Models (LLMs) impress with ever-improving language skills. However, it’s a long way from a raw LLM to a production-ready application, especially when domain knowledge is needed. This talk is your first step to unleashing the full power of LLMs for your business case.
This session gives an introduction to the three most common methods for customizing LLMs: prompt engineering, retrieval augmented generation (RAG) and fine-tuning.
We kick things off with a look under the hood of today’s LLMs and discover their inherent limitations. For our example use case, a question-answering digital doppelgänger, we tune LLaMa 3 one customization at a time to approach a useful product. We show how you can use prompt engineering to drastically improve the quality of your LLM answers, how RAG enhances the LLM with domain knowledge and how you can give your LLM the finishing touch with fine-tuning. Throughout this, we sprinkle in technical insights and live demos to understand the strengths and weaknesses of each technique.