Lag-Llama: A Foundation Model for Probabilistic Time Series Forecasting

Résumé

In this work, we present Lag-Llama, a general-purpose probabilistic time series forecasting model trained on a large collection of time series data. The model has good zero-shot prediction capabilities allowing for its use as a pre-trained time series forecasting model and we show that the model’s zero-shot learning performance improves as the model scale increases.