Digital Event Horizon
IBM has released the Granite Time Series PatchTST-FM-r2 model, a state-of-the-art time series forecasting solution with commercial-friendly licensing. The model boasts strong zero-shot performance and is designed to be highly scalable and flexible, making it an exciting development in the field of time series forecasting.
The Granite Time Series PatchTST-FM-r2 model boasts state-of-the-art performance in time series forecasting. The model provides a high-performance, zero-shot forecasting solution for a wide range of time series applications. The updated architecture incorporates conformer blocks, combining multi-head self-attention with temporal convolution. The model has been extensively benchmarked on the GIFT-Eval leaderboard, achieving strong zero-shot performance. The model is available through the Hugging Face Hub and the Granite-TSFM repository with a permissive, commercial-friendly open-source license.
IBM has recently released its latest creation, the Granite Time Series PatchTST-FM-r2 model, which boasts a state-of-the-art (SOTA) performance in time series forecasting. This latest addition to the Granite TSFM family is designed to provide a high-performance, zero-shot forecasting solution for a wide range of time series applications. The model is built on top of the patch-based representation that made the PatchTST family effective, with an updated architecture that incorporates conformer blocks, which combine multi-head self-attention with temporal convolution.
The Granite Time Series PatchTST-FM-r2 model is the result of significant improvements over its predecessor, PatchTST-FM-r1. The updated architecture includes a larger pretraining corpus, probabilistic forecasting, support for imputation of missing values, and strong zero-shot performance. The model is trained on a documented pretraining corpus consisting of four sources: selected datasets from GiftEvalPretrain; custom synthetic data based on KernelSynth with modified periodic kernels and limited augmentation; a TSMixup corpus generated using the approach described by Chronos but restricted to datasets outside the GIFT-Eval evaluation set; and approximately 500,000 synthetic CauKer sequences, each of length 4,096.
One of the key features of the Granite Time Series PatchTST-FM-r2 model is its permissive, commercial-friendly open-source license. The model is dual-licensed under Apache 2.0 and OpenMDW 1.0, which provides broad, permissive rights to use, modify, and distribute the models. This allows organizations, researchers, and developers to build on the technology with confidence, knowing that they have the freedom to use the model for commercial purposes.
The Granite Time Series PatchTST-FM-r2 model is designed to be highly scalable and flexible, with the ability to handle a wide range of time series applications. The model provides both point forecasts and quantile outputs for forecasting distributions and uncertainty intervals. It also supports very long contexts of up to 8,192 steps and flexible forecast lengths. The model backbone is constructed from conformer blocks that combine multi-head self-attention with temporal convolution to capture long- and short-range temporal structure.
The Granite Time Series PatchTST-FM-r2 model has been extensively benchmarked on the GIFT-Eval leaderboard, where it has achieved strong zero-shot performance. The model ranks #2 overall among replicable, zero-shot models and is the highest-performing model in the same category among models with permissive, commercial-friendly licensing.
IBM has made the Granite Time Series PatchTST-FM-r2 model available through the Hugging Face Hub, making it easy for developers and researchers to try the model on their own data. The model is also available through the Granite-TSFM repository, which is backward-compatible with PatchTST-FM-r1 checkpoints.
The Granite Time Series PatchTST-FM-r2 model is an exciting development in the field of time series forecasting, providing a high-performance, zero-shot forecasting solution that is both scalable and flexible. Its permissive, commercial-friendly open-source license also provides a significant advantage for organizations and researchers who need to use the model for commercial purposes.
Related Information:
https://www.digitaleventhorizon.com/articles/IBM-Unveils-State-of-the-Art-Granite-Time-Series-PatchTST-FM-r2-Model-with-Commercial-Friendly-License-deh.shtml
https://huggingface.co/blog/ibm-research/ibm-releases-sota-granite-time-series
Published: Wed Sep 9 10:59:25 2026 by llama3.2 3B Q4_K_M
