Abstract
Multi-modal large language models (MLLMs) have enabled numerous advances inunderstanding and reasoning in domains like vision, but we have not yet seenthis broad success for time-series. Although prior works on time-series MLLMshave shown promising performance in time-series forecasting, very few worksshow how an LLM could be used for time-series reasoning in natural language. Wepropose a novel multi-modal time-series LLM approach that learns generalizableinformation across various domains with powerful zero-shot performance. First,we train a lightweight time-series encoder on top of an LLM to directly extracttime-series information. Then, we fine-tune our model with chain-of-thoughtaugmented time-series tasks to encourage the model to generate reasoning paths.We show that our model learns a latent representation that reflects specifictime-series features (e.g. slope, frequency), as well as outperforming GPT-4oon a set of zero-shot reasoning tasks on a variety of domains.