205
+340
// summary
TimesFM is a decoder-only foundation model for time series forecasting developed by Google Research. The latest 2.5 version significantly improves model performance by reducing the parameter count to 200M and extending the context length to 16k. The model supports various backend configurations and provides a flexible forecasting interface to meet different application requirements.
// use cases
01
Supports a context length of up to 16k for processing complex time series data
02
Provides an optional 30M quantile head for continuous quantile forecasting
03
Supports covariate input via XReg to enhance forecasting accuracy