Better not Bigger: Distilling LLMs into Specialized Models

Explore the challenges of deploying Large Language Models (LLMs) and discover two innovative distillation solutions in this 17-minute talk by Jason Fries, a research scientist at Snorkel AI and Stanford University. Learn about the "distilling step-by-step" approach, a collaboration between Snorkel AI and Google Research, which prompts LLMs to provide answers along with their reasoning, enabling data scientists to train smaller models with similar performance using less data. Discover how the Snorkel Flow data development platform allows users to distill expertise from multiple LLMs into deployable, small-format models. Gain insights into cutting-edge techniques for creating specialized, efficient AI models from larger language models.

Better not Bigger: Distilling LLMs into Specialized Models