chronos-forecasting
Pretrained transformer foundation models for zero-shot time series forecasting across univariate and multivariate data.
Tokenize historical observations to generate probabilistic time series forecasts with foundation models.
Chronos is a Python library from Amazon Science that adapts transformer architectures into zero-shot time series forecasters. It converts observations into tokens to predict future trajectories without task-specific retraining. Data scientists and ML engineers seeking probabilistic forecasts without building custom training pipelines should use it.
What it is
The tool provides Python pipelines to execute inference on the Chronos family of foundation models, including Chronos-2 and Chronos-Bolt. It accepts pandas DataFrames containing historical targets and optional exogenous covariates, returning quantile prediction intervals. You can run inference locally on CPU or GPU hardware, or deploy endpoints to AWS SageMaker via AutoGluon-Cloud.
Who it's for
It suits data science teams requiring zero-shot probabilistic forecasting across high-volume series without manual model tuning. It is ideal if you already work within PyTorch, Hugging Face, or AWS cloud environments. It is less suited for constrained edge hardware that cannot comfortably run transformer models.
Key features
Zero-shot multivariate forecasting
Predict future time series trajectories using historical context and exogenous covariates without dataset retraining.
Patch-based Chronos-Bolt models
Chunk context into multi-observation patches for direct multi-step forecasting with lower memory usage.
Quantile probabilistic outputs
Generate prediction intervals across configurable quantile levels to directly quantify prediction uncertainty.
Pandas DataFrame integration
Pass standard tabular time series datasets directly into the pipeline prediction functions.
Up and running in 3 steps
Our honest verdict
What's great
- Delivers zero-shot forecasts without requiring custom model training pipelines on new datasets.
- Supports exogenous covariates alongside target time series in Chronos-2.
- Offers Chronos-Bolt variants designed for fast inference and reduced memory footprints.
- Integrates natively with Hugging Face transformers and Amazon SageMaker deployment workflows.
Where it falls short
- Requires dedicated GPU resources for low-latency or high-throughput batch inference.
- Exogenous covariate support is limited to Chronos-2 rather than older model variants.
- Cloud infrastructure costs can scale rapidly when hosting persistent transformer endpoints.
Pricing & license
Chronos is free and open source under the Apache-2.0 license. Operating costs stem from your local host compute or cloud resources used when deploying models on AWS SageMaker or AutoGluon-Cloud.
Open remote AI-training roles
870 open remote AI-training roles, updated November 14, 2026; most pay $20–$170 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Can I pass exogenous variables to Chronos models?
Yes, Chronos-2 supports covariate-informed forecasting by accepting a DataFrame of future covariate values.
How does Chronos generate probabilistic forecasts?
It outputs predictions across specific quantile levels (such as 0.1, 0.5, and 0.9) to provide prediction intervals.
What hardware is needed to run Chronos?
While small models can run on CPU, PyTorch GPU acceleration (CUDA) is recommended for larger models and batch inference.
