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

What happened

NVIDIA Kumo Forecast, formerly NV-Tesseract Forecasting, combines a pretrained MOMENT-1-large Transformer backbone with optional Domain-Aware Representation and Retrieval (DARR) for multivariate time-series forecasting. It uses an approximately 340-million-parameter base encoder plus a trainable forecasting head, with horizon-specific explanations that accompany predictions. Available under Apache 2.0, the library provides a DataFrame-first API, configurable forecast horizons, and separate historical context for retrieval.

The story so far

NVIDIA has introduced Kumo Forecast, formerly called NV-Tesseract Forecasting, as an Apache 2.0–licensed library that combines a pretrained Transformer model with optional historical-pattern retrieval to forecast multiple time-series variables. The library uses the MOMENT-1-large backbone, an approximately 340-million-parameter base encoder, with a trainable forecasting head. Its optional Domain-Aware Representation and Retrieval (DARR) mode blends direct model predictions with patterns retrieved from historical data. It provides a DataFrame-first API, configurable forecast horizons, separate historical context for retrieval, and explanations tied to each forecast horizon. NVIDIA’s Factory Operations (FOX) Blueprint also includes early access to a Predictive Maintenance module that uses Kumo Anomaly and Kumo Forecast.

Keep reading

Trending now

1. 1Mistral and NVIDIA Announce Industrial Engineering AI Collaboration and Open Benchmarks
2. 2Mistral Releases Voxtral 3 Speech Models, Voice Agents API and LiveKit-Based SDK
3. 3DeepSeek Open-Sources DeepInfer and API Components, With Model and Hardware Limits
4. 4OpenAI Acquires Orchestra Bio to Add Drug Development Operations to Its Life Sciences Tools