Chronos
A family of pretrained time series forecasting models offering zero-shot univariate, multivariate, and covariate-informed forecasting, deployable through AWS SageMaker or AutoGluon.
Chronos is a package providing pretrained time series forecasting models. Its latest version, Chronos-2, offers zero-shot support for univariate, multivariate, and covariate-informed forecasting tasks, and reports state-of-the-art zero-shot performance across benchmarks including fev-bench and GIFT-Eval, with the largest gains on tasks that include exogenous features.
Chronos-Bolt is an earlier, patch-based variant that chunks historical context into patches fed to an encoder, whose decoder then generates quantile forecasts for multiple future steps directly, making it up to 250 times faster and 20 times more memory-efficient than the original Chronos models of the same size. Deployment guides cover running Chronos-2 on AWS through AutoGluon-Cloud or Amazon SageMaker JumpStart for real-time, serverless, or batch inference, and a companion package called fev provides lightweight benchmarking of forecasting models built on the Hugging Face datasets library.