chronos-forecasting
Chronos: Pretrained Models for Time Series Forecasting
What is chronos-forecasting?
Chronos: Pretrained Models for Time Series Forecasting
Overview of chronos-forecasting
What it is
This package provides an interface to the Chronos family of pretrained time series forecasting models. The supported model types are Chronos-2, Chronos-Bolt, and the original Chronos. Chronos-2 offers zero-shot support for univariate, multivariate, and covariate-informed forecasting tasks. Chronos-Bolt is a patch-based variant that is up to 250 times faster and 20 times more memory-efficient than the original Chronos models of the same size. The original Chronos is based on language model architectures, transforming time series into tokens via scaling and quantization. The project is licensed under Apache-2.0.
Key features
- Zero-shot support for univariate, multivariate, and covariate-informed forecasting tasks (Chronos-2).
- Chronos-2 achieves the best performance on fev-bench, GIFT-Eval, and Chronos Benchmark II amongst pretrained models.
- Chronos-Bolt models are up to 250 times faster and 20 times more memory-efficient than original Chronos models of the same size.
- Available model sizes range from 8M to 710M parameters across the Chronos-2, Chronos-Bolt, and Chronos families.
- The package includes a
fevlightweight benchmarking package based on Hugging Face datasets. - Deployment options to AWS via AutoGluon-Cloud (recommended) or SageMaker JumpStart.
Use cases
- Time series forecasting with covariates (exogenous features).
- Probabilistic forecasting with quantile outputs.
- Benchmarking forecasting models using the
fevpackage. - Production deployment on AWS with real-time, serverless, or batch inference.
- Quick experimentation via provided example notebooks (Chronos-2 Quick Start, deployment guides).
Getting started
Install the package: sh pip install chronos-forecasting For a minimal forecasting example using Chronos-2, see the repository README.
License and pricing
Apache-2.0 License. The software is open source and free to use.
Based on the project's public README and repository metadata.