Explore how Snorkel leverages Ray to build interactive enterprise machine learning products in this 33-minute talk. Discover the challenges of creating low-latency ML solutions for diverse enterprise environments, from on-premises setups with limited resources to large-scale datasets requiring interactive processing. Learn about the architectural redesign that enables Snorkel to meet increasing data and model scale demands while maintaining product performance. Dive into distributed data/task parallelism for resource-constrained customers, scalable in-memory processing for resource-abundant clients, and the unified architecture that accommodates both approaches. Gain insights into lessons learned from using Ray to develop high-performance ML systems, and understand how this redesign powers Snorkel's flagship enterprise product, Snorkel Flow. Access the accompanying slide deck for a visual guide to the concepts discussed.
Overview
Syllabus
How Snorkel Builds Interactive Enterprise ML Products Using Ray
Taught by
Anyscale