google-research/timesfm

Health v2 · maintenance only

Adoption not part of the score

34487 stars · 3330 forks observed · 2026-10-15

What it is AI-extracted, prompt v1, taxonomy v1, 2026-10-05, confidence 0.97

TimesFM is a pretrained decoder‑only foundation model for time‑series forecasting released by Google Research. It provides zero‑shot forecasting for univariate and multivariate series with support for dynamic covariates.

Use cases

• forecast future sales demand from historical sales data
• predict energy consumption for a utility grid using past usage patterns
• generate inventory level forecasts for retail supply chain planning
• estimate financial market trends based on multivariate economic indicators
• forecast patient vital signs in healthcare monitoring systems
• predict equipment failure rates in manufacturing using sensor time‑series

When to choose

• You need high‑accuracy zero‑shot forecasting without task‑specific model training.
• Your data includes multiple correlated time‑series and dynamic covariates.
• You prefer a ready‑to‑use Python library with pretrained weights and integration with BigQuery ML or Vertex AI.

When to avoid

• You have strict non‑commercial licensing requirements for the pretrained weights.
• Your deployment environment does not support Python or the required GPU resources.
• You need interpretable models or fine‑grained control over model architecture.

Facets

library · maturity active

machine-learning artificial-intelligence machine-learning data-science python foundation-model time-series forecasting pretrained decoder-only multivariate

Member repositories

For agents

markdown · JSON · MCP: product_card(name="google-research/timesfm")