Granite-TimeSeries-TTM-R1

Granite-TimeSeries-TTM-R1 is the entry point to IBM's TinyTimeMixer line, and this catalog entry links to IBM's official TTM-R1 checkpoint. TinyTimeMixer is a compact forecasting architecture built for fast zero-shot and few-shot forecasting on standard public benchmarks. Rather than ship a single universal dense model, the family specializes checkpoints for specific context and prediction lengths, which is what keeps each one small and quick to serve. At 805K parameters it is the smallest and most deployment-friendly IBM checkpoint in the model landscape.

The TinyTimeMixer architecture is built to run efficiently on CPUs and lightweight inference paths rather than to maximize raw capacity. R1 was pretrained on public Monash forecasting datasets — Australian Electricity and Weather, Bitcoin, KDD Cup 2018, London Smart Meters, Saugeen River Flow, Solar, US Births, and wind datasets — and IBM states R1 used about 250M public training samples, giving it solid zero-shot and multivariate coverage across those domains.