Machine Learning Infrastructure at Meta Scale
Explore the challenges and solutions in scaling machine learning infrastructure at Meta in this 37-minute conference talk from MLOps World: Machine Learning in Production. Gain insights from Shivam Bharuka, Senior AI Infra Engineer at Meta, as he shares his experience in supporting large-scale ranking and recommendation models serving over a billion users. Discover how Meta reimagined its entire AI Infrastructure stack to accommodate rapidly growing machine learning models. Learn about the development of specialized hardware using powerful GPUs and network devices, as well as the design of optimized distributed training algorithms using PyTorch. Understand the approach taken to redesign and scale the stack, addressing performance, reliability, and efficiency concerns in machine learning training infrastructure.