Model Overview
Chronos-Bolt is an improved variant of the original Chronos models (from the paper “Chronos: Learning the Language of Time Series”). It is designed for time series forecasting using the same discrete-token approach as Chronos (scaling + quantization → transformer language model). Compared to Chronos, Bolt is optimized for speed and efficiency: It replaces the autoregressive decoding (one token at a time) with a non-autoregressive “parallel” prediction mechanism. This allows Chronos-Bolt to generate full forecast horizons in a single forward pass, making inference orders of magnitude faster.
Key Features
- Efficient Design: 80x faster than Chronos-Tiny during inference
- Lightweight: Smaller model size for edge deployment
- Fast Inference: One-shot prediction to replace autoregressive decoding
