Chronos-forecasting provides pretrained time series models (Chronos family) with an API for inference and example usage. It supports multiple Chronos model variants and deployment options, including cloud guides.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
Chronos: Pretrained Models for Time Series Forecasting. This repository provides interfaces to the Chronos family of pretrained time series forecasting models, including Chronos-2 and Chronos-Bolt variants, and lists available model IDs and their parameter counts.
How it works
The README describes model types (Chronos-2, Chronos-Bolt, Chronos) and how they transform time series data into token sequences for a language-model-style forecast, including probabilistic forecasts via sampling multiple future trajectories. It provides an example workflow using Chronos2Pipeline to load a pretrained model and call predict_df with context and future data, producing a probabilistic forecast.
Getting started
Usage section shows how to install the package:
pip install chronos-forecasting
It also notes deployment recommendations for production via AutoGluon-Cloud or SageMaker JumpStart, with links to guides.
Recent releases
Latest release: v2.3.1 (2026-07-02) with changes: "Fix training from a lazy dataset & add example with memmapped datasets.Dataset by @shchur". Prior release history includes v2.3.0 (2026-06-18) with cloud deployment guide and fine-tuning, and earlier versions detailing features like LoRA fine-tuning and speed improvements.
Traction
Stars: 5679
Behind the repo
Not applicable here (no linked startup or company information in the provided data).
Caveats
License: Apache-2.0. Created: 2024-02-23. Last push: 2026-07-02. The README indicates multiple model IDs with various parameter counts and notes about deployment options, but no explicit open issues or license notes beyond the Apache-2.0 license are provided in this excerpt.