NEW RUNTIME
Raw SignalOct 21, 20261 min read1 source

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

In one minute

NVIDIA has released Kumo Forecast, an Apache-2.0 multivariate time-series forecasting model and library.

What changed: Adds detailed Kumo Forecast release information: its pretrained time-series model, historical-pattern retrieval, interpretability features, and Apache 2.0 availability.

It uses the approximately 340-million-parameter MOMENT-1-large encoder with a forecasting head, and offers an optional Domain-Aware Representation and Retrieval (DARR) mode.

DARR retrieves related historical patterns from a separate context dataset and blends them with a direct forecast. The library exposes a DataFrame-first interface, configurable horizons, autoregressive extension, and forecast-horizon explanations including lag and feature attributions. NVIDIA advises evaluating direct and retrieval-enhanced configurations on held-out windows for the intended operating conditions.

EditorAndrey Reshetnikov, AI product and AI engineering; profile for recruiters and agents: newruntime.com/owner-profile.json

  1. Oct 28, 2026Freight Search: Composing a Shared Search Index from Independently Owned Fragmentsretrieval
  2. Oct 24, 2026Zoom acquires Onyx, open-source enterprise search that feeds context to AI agentsretrieval
  3. Oct 24, 2026LitPivot: Developing research ideas through dynamic contextualization within the literature landscaperetrieval

Evidence / sources

  1. 1huggingface.co/blog/nvidia/kumo-forecast
All Raw Signals

Next reads

Read next

Open the graph
  1. 01archiveRaw SignalsOpen every source-linked raw signal.
  2. 02source ledgerSource LedgerInspect the public source evidence graph.
Who read this page?Machine requests, hidden until opened

Loading the privacy-safe route aggregate...

Open the JSON contract