Toto 2.0: Time Series Forecasting Enters the Scaling Era
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866913146558480384 |
|---|---|
| author | Khwaja, Emaad Lettieri, Chris Woo, Gerald Belouadah, Eden Cenac, Marc Jarry, Guillaume Paquin, Enguerrand Zhao, Xunyi Zhukov, Viktoriya Abou-Amal, Othmane Liu, Chenghao Talwalkar, Ameet Asker, David |
| author_facet | Khwaja, Emaad Lettieri, Chris Woo, Gerald Belouadah, Eden Cenac, Marc Jarry, Guillaume Paquin, Enguerrand Zhao, Xunyi Zhukov, Viktoriya Abou-Amal, Othmane Liu, Chenghao Talwalkar, Ameet Asker, David |
| contents | We show that time series foundation models scale: a single training recipe produces reliable forecast-quality improvements from 4M to 2.5B parameters. We release Toto 2.0, a family of five open-weights forecasting models trained under this recipe. The Toto 2.0 family sets a new state of the art on three forecasting benchmarks: BOOM, our observability benchmark; GIFT-Eval, the standard general-purpose benchmark; and the recent contamination-resistant TIME benchmark. This report describes our experimental results and details the design decisions behind Toto 2.0: its architecture and training recipe, training data, and the u-muP hyperparameter transfer pipeline. All five base checkpoints are released under Apache 2.0. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_20119 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Toto 2.0: Time Series Forecasting Enters the Scaling Era Khwaja, Emaad Lettieri, Chris Woo, Gerald Belouadah, Eden Cenac, Marc Jarry, Guillaume Paquin, Enguerrand Zhao, Xunyi Zhukov, Viktoriya Abou-Amal, Othmane Liu, Chenghao Talwalkar, Ameet Asker, David Machine Learning Artificial Intelligence We show that time series foundation models scale: a single training recipe produces reliable forecast-quality improvements from 4M to 2.5B parameters. We release Toto 2.0, a family of five open-weights forecasting models trained under this recipe. The Toto 2.0 family sets a new state of the art on three forecasting benchmarks: BOOM, our observability benchmark; GIFT-Eval, the standard general-purpose benchmark; and the recent contamination-resistant TIME benchmark. This report describes our experimental results and details the design decisions behind Toto 2.0: its architecture and training recipe, training data, and the u-muP hyperparameter transfer pipeline. All five base checkpoints are released under Apache 2.0. |
| title | Toto 2.0: Time Series Forecasting Enters the Scaling Era |
| topic | Machine Learning Artificial Intelligence |
| url | https://arxiv.org/abs/2605.20119 |