Search papers, labs, and topics across Lattice.
TinyCast is a novel attention-free zero-shot forecaster that utilizes 146,505 parameters to compute periodic structures in data rather than learning them, achieving a significant breakthrough in probabilistic forecasting. By employing a zero-parameter spectral detector to identify dominant periods and a dilated convolutional encoder with a block-autoregressive quantile decoder, TinyCast establishes a new size-accuracy frontier in the GIFT-Eval benchmarks. Notably, it is the only model under 1.4M parameters that emits a predictive distribution without test-data leakage, outperforming larger models by a substantial margin in terms of parameter efficiency and accuracy.
TinyCast shatters the size-accuracy barrier in zero-shot forecasting, achieving superior performance with a fraction of the parameters of its competitors.
We introduce TinyCast, an attention-free zero-shot forecaster that emits a predictive distribution from 146,505 parameters, on the premise that at this size the periodic structure of a context is worth computing rather than learning. A zero-parameter spectral detector supplies the dominant periods, the context is folded on their phase, and a dilated convolutional encoder and a block-autoregressive quantile decoder model the rest. It is smaller than every zero-shot entry on the GIFT-Eval board whose parameter count can be established. On probabilistic accuracy it defines the size-accuracy frontier. Among zero-shot entries declaring no test-data leakage it is the only one below 1.4M parameters that emits a predictive distribution, and every entry scoring better carries at least that budget. On Chronos-ZS and fev-bench every neural model ahead of it carries at least 28 times its parameters. Because the mixing path is convolutions and matrix multiplications only, it exports to static INT8 and forecasts end to end on an embedded device without per-signal fitting.
