Breaking01 / AI
TimeOmni-o1 links language reasoning to forecasts
The researchers presented a two-way architecture that feeds language-model reasoning into time-series forecasting.

Researchers presented TimeOmni-o1, a model linking time-series forecasting with language reasoning, in a paper submitted on 14 Oct.
According to the paper, the forecasting model compresses time series, numerical values recorded over time, and passes them to a language model. The language model's reasoning then feeds back into the forecasting model to guide the generation of future values.
The researchers reported building MMTR, a training dataset with annotated reasoning traces. The paper lists 500,000+ samples across 12 domains.
The paper describes limitations in both types of existing model. Language models read time series as digit tokens and struggle with long sequences, while pretrained time-series models generally cannot read or generate text. TimeOmni-o1 connects the two to use the forecasting model's pretrained capabilities.
The researchers did not disclose evaluation results for game-development tasks in the abstract.