How Datadog is Transforming Time Series Forecasting with Toto

Watch a 13-minute conference talk from Ray Summit 2024 exploring Datadog's revolutionary approach to time series forecasting through Toto, a Time Series-Optimized Transformer. Discover how this groundbreaking model processes observability metrics with unprecedented accuracy using advanced transformer architecture. Learn about Toto's extensive training process utilizing AWS P5 instances and Ray on Kubernetes, which processed over a trillion data points, with 75% derived from real-world observability scenarios. Follow along as Emaad Khwaja explains the distinctive features that set Toto apart in time series foundation models, showcasing its ability to perform accurate zero-shot predictions on new inputs while maintaining exceptional performance in general forecasting tasks. Gain valuable insights into the architectural innovations driving Toto's success and its potential impact on both observability and broader time series analysis applications.