Google releases multivariate TimesFM-3 forecasting model

Google Research has released TimesFM-3, a 330 million-parameter foundation model for forecasting multiple related time series in one forward pass. The model accepts multiple targets, historical covariates and known future covariates without task-specific fine-tuning, according to the report. Earlier TimesFM checkpoints through version 2.5 were univariate. TimesFM-3 was pretrained on more than 1 trillion real and synthetic time points and produces point and quantile forecasts. Its repository code is Apache-2.0, but the model weights use a non-commercial license and are restricted to non-production use.