Understand this story
IBM has released Granite Timeseries Ensemble on Hugging Face, offering a four-model ensemble recipe. By default, the ensemble combines four pretrained models: PatchTST-FM-r1, PatchTST-FM-r2, FlowState, and Tiny Time Mixers. Across 97 GIFT-Eval configurations, it achieves an aggregate MASE of 0.683, outperforming the best single model, PatchTST-FM-r2, at 0.685. The recommended default Probabilistic Ensemble pools quantile forecasts equally across members with no fitting required; the IQR Ensemble instead weights members dynamically based on prediction interval width. The ensemble's main value lies in delivering stable accuracy without having to choose an architecture in advance, though PatchTST-FM-r2 still retains the lowest aggregate CRPS/MWQL.
AI-generated from public reports
Latest development · Report
IBM Granite Time Series Models' Ensemble Forecasting MethodDevelopments
1 confirmed developments
IBM Granite time series ensemble recipes can be loaded from Hugging Face
Source:Hugging Face
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Oct 22
Hugging FaceFirst-party
IBM Granite Time Series Models' Ensemble Forecasting Method
IBM has released Granite Timeseries Ensemble on Hugging Face, by default combining four pretrained models—PatchTST-FM-r1, PatchTST-FM-r2, FlowState, and Tiny Time Mixers—achieving an aggregate MASE of 0.683 across 97 GIFT-Eval configurations, outperforming the best single model, PatchTST-FM-r2, at 0.685. The recommended default Probabilistic Ensemble pools quantile forecasts from members with equal weights and requires no fitting; the IQR Ensemble, by contrast, weights dynamically according to forecast interval width. The main value of the ensemble is stable accuracy without having to select an architecture in advance, but PatchTST-FM-r2 still retains the lowest aggregate CRPS/MWQL.