amazon-science/chronos-forecasting

What is it?

Key features

• 01Zero-shot forecasting on univariate, multivariate, and covariate-informed time series with Chronos-2.
• 02Probabilistic forecasts with configurable quantile levels for prediction intervals.
• 03Pandas DataFrame interface via predict_df with id, timestamp, and target columns.
• 04Models hosted on Hugging Face in sizes from 8M to 710M parameters.
• 05Deployment guides for Amazon SageMaker JumpStart and AutoGluon-Cloud.

Use cases

• Forecasting demand, sales, or electricity prices without building a dedicated model for each series.
• Adding probabilistic prediction intervals to an existing time series pipeline.
• Prototyping forecasts on data that includes known future covariates such as promotions or weather.
• Deploying a forecasting endpoint on AWS for production inference.

Quick start

1. 1

   Install the package with pip

   pip install chronos-forecasting
2. 2

   Install pandas with pyarrow support (needed to load the example data)

   pip install 'pandas[pyarrow]'
3. 3

   Load Chronos-2 and run the minimal forecasting example from the README (requires a CUDA GPU as written)

Things to know

• !The minimal example uses device_map="cuda", so it expects an NVIDIA GPU unless you change the device.
• !Production deployment is recommended on Amazon SageMaker or AutoGluon-Cloud, which implies AWS setup effort for production use.

Topics

FAQ

What is chronos-forecasting?

Pretrained Transformer-based models from Amazon for zero-shot time series forecasting, including the Chronos-2 family. Chronos is a family of pretrained time series forecasting models from Amazon, installable from PyPI as the chronos-forecasting package. Users load a model from Hugging Face and generate probabilistic forecasts on new series without training on them first.

Is chronos-forecasting free to use? What license does it have?

chronos-forecasting is open source under the Apache-2.0 license. Check the license terms before using it in commercial products.

Is chronos-forecasting still maintained?

Actively maintained. The last commit to chronos-forecasting was on Oct 5, 2026.

What can I use chronos-forecasting for?

Forecasting demand, sales, or electricity prices without building a dedicated model for each series. Adding probabilistic prediction intervals to an existing time series pipeline. Prototyping forecasts on data that includes known future covariates such as promotions or weather. Deploying a forecasting endpoint on AWS for production inference.

How do I install chronos-forecasting?

Install the package with pip: pip install chronos-forecasting Install pandas with pyarrow support (needed to load the example data): pip install 'pandas[pyarrow]' Load Chronos-2 and run the minimal forecasting example from the README (requires a CUDA GPU as written)

Where can I get chronos-forecasting?

The source code and installation instructions are on GitHub: https://github.com/amazon-science/chronos-forecasting