---
id: chronos-forecasting
version: "2.3.1"
license: Apache License Version 2.0, January 2004 http://www.apache.org/licenses/ TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION 1. Definitions. "License" shall mean the terms and conditions for… (full text in the JSON record)
license_treatment: permissive
maintenance: active
---

# chronos-forecasting — Chronos: Pretrained models for time series forecasting

License: permissive · Maintenance: active · Downloads: 667.0K/mo

## What it is and what it does

Chronos is a library that wraps pretrained language-model-based time series forecasting models. It transforms historical time series data into token sequences, uses transformer architectures to learn patterns, and generates probabilistic forecasts by sampling multiple future trajectories or computing quantile predictions. The library includes three model families: Chronos-2 (latest, 120M parameters, best zero-shot performance), Chronos-Bolt (patch-based, up to 250 times faster and 20 times more memory-efficient), and the original Chronos (T5-based, ranging from 8M to 710M parameters).

You use it by loading a pretrained model via HuggingFace, passing historical time series data as a pandas DataFrame with optional covariate columns, and calling predict_df to get quantile forecasts for future steps. The library handles the tokenization, model inference, and output formatting automatically. It's designed for zero-shot forecasting—applying a model trained on diverse time series to new domains without fine-tuning—and supports both univariate (single target column) and multivariate (multiple targets plus exogenous features) scenarios.

Use it for:

- Forecast electricity prices or demand using historical data and optional future weather or market covariates.
- Generate probabilistic predictions for stock prices or financial metrics with quantile-based confidence intervals.
- Predict website traffic, server load, or system metrics for capacity planning without retraining.
- Perform rapid prototyping of time series predictions on new datasets without labeled training data.
- Deploy production forecasting endpoints on AWS SageMaker or AutoGluon-Cloud for real-time or batch inference.

## Worth the install?

AI-flagged interpretation of the facts above — verify before relying.

Chronos provides pretrained transformer-based models for zero-shot time series forecasting, supporting univariate, multivariate, and covariate-informed tasks with quantile-based probabilistic predictions.

Yes. Chronos is actively maintained, has no known vulnerabilities, and offers a straightforward interface to state-of-the-art pretrained models. Install it if you need time series forecasting without the overhead of training your own model. The low install friction and permissive license make it a practical choice for prototyping and production use. Consider it essential if you're forecasting on new domains where labeled training data is scarce.

## Install

pip install chronos-forecasting
uv add chronos-forecasting
poetry add chronos-forecasting

## Installing chronos-forecasting

Before you install: Low install friction with a pure-Python wheel. Active maintenance with recent releases (last commit 2026-08-14, 43 days since latest version). Depends on torch, transformers, and accelerate—standard ML stack dependencies that may require significant disk space but are widely available.

License in practice: Apache License 2.0 permits commercial and derivative use with minimal restrictions. You must include a copy of the license and note any modifications, but there are no copyleft obligations or patent claims that would restrict downstream use.

Quickstart:

pip install chronos-forecasting

from chronos import Chronos2Pipeline
import pandas as pd

pipeline = Chronos2Pipeline.from_pretrained("amazon/chronos-2", device_map="cuda")
context_df = pd.read_parquet("train.parquet")
pred_df = pipeline.predict_df(context_df, prediction_length=24, quantile_levels=[0.1, 0.5, 0.9], id_column="id", timestamp_column="timestamp", target="target")

Requires Python >=3.10. torch and transformers must be installed; GPU recommended for inference speed but CPU inference is supported via device_map.

Verify before relying:

- Whether pretrained model weights are automatically downloaded on first use or require manual setup.
- Memory and disk requirements for different model sizes (8M to 710M parameters).
- Inference latency and throughput benchmarks on CPU vs. GPU hardware.
- Whether the package supports batch inference across multiple time series simultaneously.

## Package facts

- License: Apache License Version 2.0, January 2004 http://www.apache.org/licenses/ TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION 1. Definitions. "License" shall mean the terms and conditions for… (full text in the JSON record) (permissive)
- Python support: supports_current
- Install friction: low
- Maintenance: active
- Downloads: 667.0K/month (top 15,000 on PyPI)
- Known vulnerabilities: none known

## Tags

time series forecasting, pretrained forecasting models, zero-shot time series prediction, probabilistic forecasting, multivariate time series, transformer forecasting, quantile forecasts, time-series, forecasting, foundation-models

[View on SkillFed](https://skillfed.io/packages/chronos-forecasting) · [View on PyPI](https://pypi.org/project/chronos-forecasting/)