google-research/timesfm

Overview

TimesFM is a pretrained time-series foundation model released by Google Research for forecasting across a wide range of domains, accompanied by code and utilities for inference and fine-tuning.[3] It abstracts much of the complexity of building strong baselines for time-series ML, letting developers quickly plug in their data for forecasting experiments. Developers should care because it brings modern foundation-model capabilities to a traditionally niche area with a practical, reusable implementation.