✦
Summary
TimesFM-3 is a time-series foundation model that forecasts multiple related signals without task-specific fine-tuning. It can combine target histories with auxiliary variables such as past observations and known future events, supporting forecasting problems where relationships between series carry useful predictive information.
Key Features
The model jointly predicts multiple related time series.
Zero-shot forecasting works without dataset-specific fine-tuning.
Past covariates provide additional observations known only historically.
Past-future covariates incorporate known upcoming signals such as scheduled promotions.
The model supports quantile forecasts in addition to point predictions.
TimesFM-3 contains approximately 330 million parameters.
The model generates its forecast in a single non-autoregressive forward pass.
The repository specifies a separate noncommercial license for TimesFM 3.0 weights.
About TimesFM-3
The 330M-parameter model uses patched time-series inputs and alternating temporal and cross-series attention. Historical signals remain causally constrained, while known future covariates use a lookahead representation. A non-autoregressive decoding strategy produces forecasts in a single forward pass rather than generating each future patch sequentially.
TimesFM-3 is useful for research on demand, sensors, operations, and other multivariate forecasting tasks. It supplies point and quantile predictions and is pretrained on more than one trillion time points. The public code and pretrained weights have different licensing terms; the 3.0 weights are restricted to noncommercial use.