Struct TimesFMModel 

pub struct TimesFMModel { /* private fields */ }

TimesFM 2.5 200M’s architecture is fixed (see the module-level consts above) — there is no config.json and hence no InferConfig/builder here, unlike every other model in the workspace. TimesFMModel::load takes just the GGUF path.

Implementations

impl TimesFMModel

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

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

Forecast a univariate time series.

Returns [N_OUTPUTS][prediction_length] where index 0 is the point forecast and indices 1-9 are quantile forecasts (q0.1 … q0.9).

Trait Implementations

impl Forecaster for TimesFMModel

type Config = ()

No config.json exists for this model — architecture is fixed at compile time.

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

TimesFM’s forecast() is univariate-only and always treats the context as fully observed; mask is unused.

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