LTSM-Bundle: A Toolbox and Benchmark on Large Language Models for Time Series Forecasting

Time Series Forecasting (TSF) has long been a challenge in time series analysis. Inspired by the success of Large Language Models (LLMs), researchers are now developing Large Time Series Models (LTSMs), universal transformerbased models that use autoregressive prediction to improve TSF. However, training LTSMs on heterogeneous time series data poses unique challenges, including diverse frequencies, dimensions, scalability, and patterns across datasets. Recent efforts have studied and evaluated various design choices aimed at enhancing LTSM training and generalization capabilities. Despite progress in both paradigms, there is no unified framework for systematically evaluating models and design choices across them. However, these design choices are typically studied and evaluated in isolation and are not compared collectively. In this work, we introduce LTSM-Bundle, a comprehensive toolbox and benchmark for training LTSMs, spanning pre-processing techniques, model configurations, and dataset configurations. We modularize and benchmark LTSMs across multiple dimensions, including prompting strategies, tokenization approaches, training paradigms, base model selection, data quantity, and dataset diversity. By combining the most effective design choices, the combination achieves state-of-the-art zero-shot and few-shot performance while providing a reproducible foundation for evaluating both traditional LSF models and emerging LTSMs. Our source code is available at https: //github.com/datamllab/ltsm