Cutting Large Language Models Down to Size with Dr. Nathan Wycoff
Abstract: Large Language Models such as ChatGPT have captured the zeitgeist and portend a new wave of digital disruption across all sectors. In this session, we will start with a close look at what large language models are and what makes these transformer-based architectures different from previous probabilistic language models. We will accomplish this by building our own tiny language model. Then we will discuss strategies for exploiting existing LLMs as is, using them as sophisticated text embeddings and prompt answerers. We will conclude by fine-tuning LLMs for a specific task, highlighting their ability to obtain higher predictive accuracy for text-based application than traditional machine learning methods.Â
Bio: Dr. Nathan Wycoff is a Data Science Fellow at Georgetown’s Massive Data Institute working on the Forced Migration and Data Privacy teams. Dr. Wycoff received his PhD in statistics from Virginia Tech.