Optimizing LLM Training with Airbnb's Next-Gen ML Platform

Explore how Airbnb optimizes large language model (LLM) training using their next-generation machine learning platform in this 28-minute conference talk. Discover how Airbnb's ML Platform team leveraged Ray to enhance their existing infrastructure, enabling internal teams to effectively utilize open-source LLMs across multiple use cases. Learn about the scalability challenges posed by LLMs and how Airbnb overcame them. Gain insights into the improvements in developer productivity, performance enhancements, and better utilization of A100 GPUs on AWS for training workloads. Compare before-and-after states, examine real-world benchmarks demonstrating cloud efficiency, and contrast Sagemaker with Ray. Understand the long-term efficiency and cost savings achieved for strategic LLM applications, and explore the various applications enabled by this work.