TimesFM is a pretrained decoder-only time-series foundation model from Google Research that provides zero-shot forecasting without per-task model tuning. It supports univariate and multivariate forecasting with covariates and ships PyTorch and MLX inference libraries.
Project overview
It replaces per-task model tuning with a single pretrained foundation model that performs zero-shot multivariate forecasting with both past-only and past-and-future covariates, and it is also available through Google BigQuery ML for commercial and production use.
Project type
Model Runtime · Data Processing
Use cases
Data Analysis
Deployment
Refer to project documentation
License
Apache-2.0
Best for
Developers and AI engineers who need time-series forecasts without per-task model tuning and are comfortable writing Python code to configure checkpoints and run inference.
Key capabilities
Forecasts 1D time series of variable context lengths, returning point predictions and quantiles.
Forecasts multi-channel series with native past-only and past-and-future covariate support without per-task tuning.
A fine-tuning example using HuggingFace Transformers with PEFT (LoRA) is provided in the examples directory.
Limitations and risks
This open-source version is not an officially supported Google product, so support expectations should be set accordingly before adoption.
Getting started
Install with pip install timesfm[torch]; setup is medium difficulty because it requires Python code and checkpoint configuration. Model checkpoints, such as google/timesfm-3.0-pytorch, are downloaded from Hugging Face.
Evidence and sources
README: TimesFM (Time Series Foundation Model) is a pretrained time-series foundation model developed by Google Research for time-series forecasting.
README: pip install timesfm[torch]
README: [A decoder-only foundation model for time-series forecasting](https://arxiv.org/abs/2310.10688), ICML 2024.
README: TimesFM 3.0 introduces native **multivariate time-series forecasting**, flexible **covariate support** (both past-only and past-and-future covariates), superior zero-shot generali…
README: Commercial and production use of TimesFM 3.0 is **fully permitted** through authorized Google Cloud services, including [BigQuery ML](https://docs.cloud.google.com/bigquery/docs/t…