Struct MoiraiModel 

pub struct MoiraiModel { /* private fields */ }

Implementations

impl MoiraiModel

pub fn load(gguf_path: &Path, config: MoiraiConfig) -> Result<Self>

pub fn forecast(&self, context: &[f32], horizon: usize) -> Result<Vec<f32>>

Zero-shot forecasting: embed context + masked future, single encoder pass.

Trait Implementations

impl Forecaster for MoiraiModel

fn forecast( &self, context: &[Vec<f32>], _mask: &[Vec<bool>], horizon: usize, ) -> Result<QuantileMatrix>

Moirai is channel-independent: each variate is forecast separately (no cross-variate attention), point-forecast only, so mask is unused. Returns a single “quantile” row (the point forecast) with one column per variate. Variates are forecast in parallel via rayon — each is a fully independent forward pass through the same read-only weights (the only shared mutable state, rope_cache, is behind a Mutex).

type Config = MoiraiConfig

fn load(gguf_path: &Path, config: MoiraiConfig) -> Result<Self>