Building Instacart's ML Training Platform on Ray

Explore how Instacart is building its next-generation ML Training Platform on Ray in this 30-minute conference talk. Discover the standard training runtime developed for broad ML use cases across Instacart, along with capabilities for advanced users to fine-tune distributed deep learning models using Ray AIR. Examine the end-to-end system built on KubeRay for workflow orchestration, and learn about the standard runtime's deep integration with Ray Data, Ray Train, and Ray Tune to leverage distributed computation capabilities throughout all training lifecycles. Delve into a custom use case featuring Instacart's personalization ranker, built on Ray AIR with complex data preprocessing steps. Gain insights into fine-tuning the p13n ranker model to achieve higher training throughput and optimizations through streaming executor. Access the accompanying slide deck for visual references and additional information.