amazon-science/chronos-forecasting

Based on the README content, here's a technical summary: Supports zero-shot univariate, multivariate, and covariate-informed forecasting through transformer-based architectures (T5 encoder-decoder or patch-based variants), with quantile prediction for probabilistic forecasts. Chronos-2 handles exogenous features directly, while Chronos-Bolt uses patched input chunks and direct multi-step decoding for 250x faster inference. Models are distributed via Hugging Face and integrate with AWS SageMaker for production deployment, with the Python package providing pandas-compatible DataFrame APIs for batch prediction.

4,920 stars. Used by 2 other packages. Actively maintained with 1 commit in the last 30 days. Available on PyPI.