Chronos Bolt Base

Chronos Bolt Base is a time series foundation model hosted on Hugging Face. Developed by Amazon, it supports probabilistic forecasting through a transformer-based approach and belongs to the class of pretrained time-series foundation models.

The model produces quantile predictions for future values. It accepts historical time series data as input and generates forecasts for a specified number of steps ahead, with support for user-defined quantile levels such as 0.1, 0.5, and 0.9. Integration occurs through the Chronos Python library, which provides a BaseChronosPipeline class for loading the model and running inference on dataframes. Example usage loads a CSV file containing context data and calls predict_df with parameters for prediction length, quantiles, and an item identifier column.

Delivery takes place as a downloadable model on the Hugging Face platform under the identifier amazon/chronos-bolt-base. It is provided in Safetensors format and can be loaded with device mapping for GPU acceleration. The model card lists associated research references including arxiv:1910.10683 and arxiv:2403.07815.

It carries an Apache-2.0 license. Access is open and free for download and use.