part three

Part 3 Using LLMs for time-series forecasting

Having explored models specifically built for time-series forecasting, we wonder whether we can use LLMs directly. After all, LLMs are large models that are trained for the analogous task of generating a sequence of words based on an input sentence.

To that end, in chapter 8, we use Flan-T5 and Llama for time-series forecasting, using prompting techniques such as few-shot and chain-of-thought prompting. We also use these models for anomaly detection and realize that although they work, they may not be the best solution for all use cases. In chapter 9, we explore Time-LLM, which greatly improves the performance of LLMs in time-series forecasting by reprogramming some of their components.