Thoth: Mid-Training Bridges LLMs to Time Series Understanding
Jiafeng Lin, Yuxuan Wang, Jialong Wu, Huakun Luo, Zhongyi Pei, Jianmin Wang · 2026
Read Thoth: Mid-Training Bridges LLMs to Time Series Understanding by Jiafeng Lin, Yuxuan Wang, Jialong Wu, Huakun Luo, Zhongyi Pei, Jianmin Wang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-01042.txt.
Description
Lin and five colleagues bridge language models and time series with Thoth, the first family of mid-trained models for general time-series understanding. Mid-training aligns time series with language in a task-agnostic way. It is powered by Book-of-Thoth, a corpus supporting generation in both directions between time series and text. A new KnoTS benchmark tests combined reasoning about time patterns and field knowledge, and Thoth beats its base model and advanced rivals, even when fine-tuning data is scarce.
Catalog record
| CREATOR | Jiafeng Lin, Yuxuan Wang, Jialong Wu, Huakun Luo, Zhongyi Pei, Jianmin Wang |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 50760 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.01042