Description
timesfm-forecasting is an agent skill for zero-shot time-series forecasting with Google's TimesFM. It covers preparing regular-grid CSV data, producing quantile forecasts, using XReg covariates, and held-out evaluation. The skill begins by making you choose the model and its interface, because the Python package version and the model checkpoint version are separate. For TimesFM 2.5 with 200 million parameters, the interface gives a mean and nine deciles, and its weights are Apache-2.0. The newer TimesFM 3.0, with about 330 million parameters, gives nine deciles and has downloaded weights restricted to non-commercial, non-production use. The skill keeps 2.5 as its default local workflow because its Apache-licensed weights have different usage terms. For 3.0 work on multivariate targets, past-only covariates or Apple MLX, it points to a separate reference file to read before adapting code. A Google Cloud entitlement does not change the licence of downloaded weights, so check the upstream licence notice before commercial use. Use it when you need a forecast for a regular time series without training a model on your own data.…
Highlights
- Zero-shot time-series forecasting
- Quantile forecasts
- XReg covariates and held-out checks
- Regular-grid CSV preparation
- Licence differences between model versions
Ratings & Reviews
No ratings yet
Be the first to rate timesfm-forecasting.
People also visit
pytorch-patterns
PyTorch deep learning patterns and best practices for building robust
timesfm-forecasting FAQ
What does it forecast?
Time series, zero-shot, with quantile forecasts and optional covariates.
Which model is the default?
TimesFM 2.5, because its weights are Apache-2.0 licensed.
Can I use TimesFM 3.0 commercially?
The skill says downloaded 3.0 weights are restricted to non-commercial, non-production use. Check the upstream licence notice.
Does a Google Cloud entitlement change the licence?
No. The skill says it does not change the licence of downloaded weights.
How do I install it?
Run npx skills add https://github.com/k-dense-ai/scientific-agent-skills --skill timesfm-forecasting.
Is it checked for security?
Its security audits show passes from Gen Agent Trust Hub, Socket and Snyk.
What are the best timesfm-forecasting alternatives?
Popular alternatives to timesfm-forecasting include aeon, pytorch-patterns and deepchem. See all timesfm-forecasting alternatives