NVIDIA releases Kumo Forecast, a multi-variate time-series model combining pretrained representations with historical pattern retrieval

NVIDIA Kumo Forecast is a multi-variate time-series forecasting model that combines a pretrained Transformer backbone with Domain-Aware Representation and Retrieval (DARR). Built on the MOMENT-1-large encoder with approximately 340 million parameters, it offers configurable forecast horizons and an optional retrieval pathway that blends direct model predictions with patterns retrieved from historical data. The library provides a DataFrame-first API and an interpretability framework producing horizon-specific explanations.