See More, Forecast Better and Faster: Enhancing Time Series Foundation Models via Inference-Time Plug-and-Play Downsampling

Time series foundation models (TSFMs) have demonstrated impressive generalization capabilities across diverse domains. However, they face significant challenges in long-term and ultra long-term forecasting. These challenges primarily arise from scalability limitations when TSFMs process extensive sequence lengths. To address this, we propose SPRINT, a training-free plug-and-play framework designed to empower TSFMs to see more, forecast better and faster during inference. The core idea is to perform forecasting in a downsampled-resolution space, enabling an extended look-back window with reduced computational costs. To avoid information loss and resolution mismatch caused by downsampling, SPRINT decomposes time series into trend and seasonal components, processing them separately. It predicts the low-frequency trend via a Resolution Interpolation workflow within the downsampled space, while preserving high-frequency details through a Pattern Replication mechanism for seasonality. Extensive experiments show that SPRINT achieves a significant improvement, increasing accuracy by 19\% while enhancing efficiency with a reduction of max memory usage by 6.4$\times$ and inference time by 16.9$\times$ compared to state-of-the-art TSFMs.

Many important systems, such as energy grids, weather stations, and online services, produce streams of measurements over time. Predicting far into the future from these streams is useful, but large forecasting models become slow and memory-hungry when they must read very long histories or make very long forecasts. This paper introduces SPRINT, a method that helps existing forecasting models handle longer sequences without retraining them. SPRINT separates a signal into repeated patterns, such as daily or weekly cycles, and a smoother long-term trend. It preserves the repeated patterns directly, asks the model to forecast a compressed version of the smoother trend, and then expands the result back to the original resolution. Across many datasets and models, this approach improves forecast accuracy while greatly reducing memory use and prediction time, making long-range forecasting more practical for real-world high-frequency data.