SettledModels
IBM releases Granite Timeseries Ensemble, combining four forecasters without retraining
IBM Research released a recipe that pools four Granite forecasters, from 1.41M to 0.4B parameters, into one point and quantile forecast with no retraining. On GIFT-Eval's 97 configurations it scores MASE 0.683 and ranks third among eligible systems. IBM's own numbers; some member checkpoints carry non-commercial terms.
- HEAT
- OUTLETS
- 0
- FIRST SEEN
- LAST SIGNAL
Primary source
Hugging Facehuggingface.co/blog/ibm-research/granite-timeseries-ensemble-r1Coverage · 1 article, oldest first
Thursday, October 22, 2026
Hugging FacePrimary
Ensemble forecasting with IBM Granite time-series models