IBM releases SOTA Granite Time Series PatchTST-FM-r2 model with commercial-friendly license

IBM has released Granite Time Series PatchTST-FM-r2, a new foundation model for zero-shot forecasting of time-series data. The approximately 385M parameter model features an updated architecture using conformer blocks that combine multi-head self-attention with temporal convolution. It demonstrates strong zero-shot performance on the GIFT-Eval benchmark, ranking as the top-performing model with a permissive, commercial-friendly license. The model weights, architecture, and inference code are all openly available on Hugging Face under an Apache 2.0 license.