timesfm is a pretrained decoder-only foundation model for time-series forecasting.
Project overview
By taking a decoder-only foundation-model approach to time-series forecasting, timesfm offers a pretrained architecture that can be applied directly to forecasting tasks, reducing the need to train a model from scratch.
Project type
Model Runtime
Use cases
Data Analysis
Deployment
Refer to project documentation
License
Apache-2.0
Best for
Those who need a pretrained foundation model for time-series forecasting tasks.
Key capabilities
Provides a decoder-only foundation model for generating time-series forecasts.
Includes covariate support through XReg.
Supports fine-tuning using HuggingFace Transformers and PEFT (LoRA).
Offers a Flax version of the model for faster inference.
Limitations and risks
This open version is not an officially supported Google product.
Getting started
Install the package quickly using pip install timesfm.
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: 1. ✅ Flax version of the model for faster inference.
README: Added back the covariate support through XReg for TimesFM 2.5.
README: Added fine-tuning example using HuggingFace Transformers + PEFT (LoRA)
README: * [BigQuery ML](https://cloud.google.com/bigquery/docs/timesfm-model): Enterprise level SQL queries for scalability and reliability. * [Google Sheets](https://workspaceupdates.goo…