From the model card
Fields below are copied from the tags and counters on the HuggingFace repository amazon/chronos-bolt-base at our last fetch. They are set by the uploader, not verified by us; rows with no tag are omitted. How this page is made.
- Publisher (HF namespace)
- amazon
- Pipeline tag
- time-series-forecasting
- Weight formats
- safetensors
- License tag
apache-2.0— read the license file in the repo before relying on it- Papers cited
- arXiv:1910.10683, arXiv:2403.07815
- Downloads (HF counter at last fetch)
- 1,166,830
- Likes (HF counter at last fetch)
- 92
- Model card
- https://huggingface.co/amazon/chronos-bolt-base
Use cases
- Zero-shot time-series forecasting across diverse domains
- Forecasting pipeline benchmarks against traditional statistical methods
- Demand planning prototyping without per-dataset training
- Multi-horizon forecasting with variable horizon lengths
- Ensemble component in combined forecasting systems
Pros
- Improved accuracy over original Chronos at equivalent size via Bolt training
- Apache 2.0 license
- Zero-shot domain transfer without fine-tuning
- T5 architecture handles variable forecast horizons
Cons
- Requires custom chronos-forecasting library code or AutoGluon for inference
- Not competitive with dataset-specific models on narrow single-domain tasks
- Token quantization introduces discretization error for continuous series
- Higher latency than classical statistical forecasting methods
- Performance on irregular or event-driven series is limited
Tags
chronos-forecastingsafetensorst5time seriesforecastingpretrained modelsfoundation modelstime series foundation modelstime-seriestime-series-forecastingarxiv:1910.10683arxiv:2403.07815license:apache-2.0region:usdeploy:sagemaker