Moirai is an innovative time series foundation model that offers versatile forecasting capabilities across various domains, frequencies, and variables. It overcomes challenges through a large-scale diverse dataset, multiple patch size projection layers, any-variate attention mechanism, and mixture distribution integration. Moirai outperforms full-shot models, demonstrating zero-shot forecasting skills in both in and out-of-distribution settings. The model tackles the inefficiency of task-specific forecasters, offering a universal solution with a pre-trained large model. This breakthrough eliminates the need for extensive computational resources and expert input, democratizing accurate forecasting across sectors like IT Operations, Sales, and Energy.
https://blog.salesforceairesearch.com/moirai/