Tiny Time Mixers
Summary
Tiny Time Mixers, or TTM, is IBM Granite’s compact pretrained forecasting model family built on TSMixer-style mixing blocks. It is notable because the released models are small, CPU-friendly, and still competitive in zero-shot and few-shot forecasting settings.
Role In The Wiki
TTM anchors the small-pretrained-model route for time-series foundation models: instead of relying on large attention-heavy backbones, it combines mixer-style architecture, adaptive patching, resolution-aware pretraining, and lightweight target adaptation.
Official Artifacts
- Granite TSFM source: https://github.com/ibm-granite/granite-tsfm
- Granite TTM r2 checkpoint: https://huggingface.co/ibm-granite/granite-timeseries-ttm-r2
Evidence
Relation To Foundation TSFM Agenda
Use the source-level agenda mappings rather than duplicating verdict rows here:
At the entity level, TTM anchors the small-pretrained-model route for time-series foundation models: instead of relying on large attention-heavy backbones, it combines mixer-style architecture, adaptive patching, resolution-aware pretraining, and lightweight target adaptation. This page should stay as the object card; source pages carry slot-level verdicts, evidence, and missing pieces.