🚀 Introducing Chronos-2: From univariate to universal forecasting
This release adds support for Chronos-2. It is a 120M-parameter time series foundation model that offers zero-shot support for univariate, multivariate, and covariate-informed forecasting tasks. Chronos-2 delivers state-of-the-art zero-shot performance across multiple benchmarks (including fev-bench and GIFT-Eval), with the largest improvements observed on tasks that include exogenous features. In head-to-head comparisons, it outperforms its predecessor, Chronos-Bolt, over 90% of times.
📌 Get started with Chronos-2: Chronos-2 Quick Start
Chronos-2 offers significant improvements in capabilities and can handle diverse forecasting scenarios not supported by earlier models.
| Capability | Chronos | Chronos-Bolt | Chronos-2 |
|---|---|---|---|
| Univariate Forecasting | ✅ | ✅ | ✅ |
| Cross-learning across items | ❌ | ❌ | ✅ |
| Multivariate Forecasting | ❌ | ❌ | ✅ |
| Past-only (real/categorical) covariates | ❌ | ❌ | ✅ |
| Known future (real/categorical) covariates | 🧩 | 🧩 | ✅ |
| Fine-tuning support | ✅ | ✅ | ✅ |
| Max. Context Length | 512 | 2048 | 8192 |
🧩 Chronos/Chronos-Bolt do not natively support future covariates, but they can be combined with external covariate regressors (see AutoGluon tutorial). This only models per-timestep effects, not effects across time. In contrast, Chronos-2 supports all covariate types natively.
Figure 1: The complete Chronos-2 pipeline. Input time series (targets and covariates) are first normalized using a robust scaling scheme, after which a time index and mask meta features are added. The resulting sequences are split into non-overlapping patches and mapped to high-dimensional embeddings via a residual network. The core transformer stack operates on these patch embeddings and produces multi-patch quantile outputs corresponding to the future patches masked out in the input. Each transformer block alternates between time and group attention layers: the time attention layer aggregates information across patches within a single time series, while the group attention layer aggregates information across all series within a group at each patch index. The figure illustrates two multivariate time series with one known covariate each, with corresponding groups highlighted in blue and red. This example is for illustration purposes only; Chronos-2 supports arbitrary numbers of targets and optional covariates.
Figure 2: Results of experiments on the fev-bench time series benchmark. The average win rate and skill score are computed with respect to the scaled quantile loss (SQL) metric, which evaluates probabilistic forecasting performance. Higher values are better for both. Chronos-2 outperforms all existing pretrained models by a substantial margin on this comprehensive benchmark, which includes univariate, multivariate, and covariate-informed forecasting tasks.
Figure 3: Chronos-2 results in univariate mode and the corresponding gains from in-context learning (ICL), shown as stacked bars on the covariates subset of fev-bench. ICL delivers large gains on tasks with covariates, demonstrating Chronos-2’s ability to effectively use covariates through ICL. Besides Chronos-2, only TabPFN-TS and COSMIC support covariates, and Chronos-2 outperforms all baselines (including TabPFN-TS and COSMIC) by a wide margin.
Figure 4: Results on the GIFT-Eval time series benchmark. The average win rate and skill score with respect to the (a) probabilistic and (b) point forecasting metrics. Higher values are better for both win rate and skill score. Chronos-2 outperforms the previously best-performing models, TimesFM-2.5 and TiRex.
What's Changed
- Add Chronos-2 by @abdulfatir in #319
- Remove ALWAYS_DOWNLOAD from CF and S3 by @abdulfatir in #322
- Use dynamic versioning and bump version by @abdulfatir in #320
- Add example notebook for Chronos-2 by @shchur in #325
- Update README for Chronos-2 by @abdulfatir in #324
- Bump version to 2.0.0 by @abdulfatir in #323
Full Changelog: v1.5.3...v2.0.0
