NVIDIA Kumo Forecast Combines Pretrained Time-Series Representations with Historical Pattern Retrieval and Interpretability

NVIDIA has introduced Kumo Forecast, a multivariate time-series forecasting model that pairs a pretrained Transformer backbone with retrieval of patterns from historical data.

• The model, formerly known as NV-Tesseract Forecasting, is built on the MOMENT-1-large backbone and uses an approximately 340-million-parameter base encoder plus a trainable forecasting head.
• Its Domain-Aware Representation and Retrieval mode adds k-nearest-neighbor retrieval from historical context, which the SDK accepts through a separate context_df argument, and developers can adjust how strongly the final prediction relies on each pathway.
• The library offers a DataFrame-first API, configurable forecast horizons, and autoregressive extension for horizons beyond the model's native capability.
• An interpretability framework produces lag and feature attributions per forecast horizon, semantic-flow and trajectory-stability diagnostics, and optional embedding Integrated Gradients without changing the underlying forecaster.
• Kumo Forecast is available under the Apache 2.0 license, with model code at NVIDIA/Kumo-TS and weights at nvidia/Kumo-Forecast.