FlowState
Tiny TSFM (~2M params) combining a state-space-model encoder with a functional-basis decoder. Inhabits a timescale-invariant coefficient space, so a single training run helps inference at every sampling rate. SOTA on GIFT-ZS and Chronos-ZS at minuscule cost. NeurIPS 2025.
FlowState is IBM Research's argument that sampling rate invariance is the missing piece in TSFM design. The encoder is a state-space model (SSM); the decoder projects onto a functional basis rather than emitting samples directly. Together they let the model operate in a continuous, timescale-invariant coefficient space — training on hourly data helps inference on 15-minute data and vice versa, all from the same checkpoint.
At roughly 2 million parameters FlowState is the smallest model in the line-up by a wide margin, yet it reports SOTA on the GIFT-ZS and Chronos-ZS benchmarks and is the only TSFM here designed from the ground up to handle arbitrary, even unseen, sampling rates without retraining. Accepted at NeurIPS 2025.
Versions on TS-Arena
Each version below corresponds to one registered model id in the leaderboard. Click through to its detail page for per-model rankings, forecasts, and history.
- FlowStateflowstate2M params…
Single released checkpoint; SSM encoder + functional-basis decoder.