Struct MultiscreenEngine 

pub struct MultiscreenEngine { /* private fields */ }

Main user-facing engine.

Implementations

impl MultiscreenEngine

pub fn new(config: MultiscreenConfig) -> Result<Self>

Creates a new engine with the given configuration.

pub fn from_weights_file(path: impl AsRef<Path>) -> Result<Self>

Creates a new engine by loading a weights file.

The config embedded in the weights file becomes the engine’s config. This is the easiest way to resume from a previously saved model.

pub fn config(&self) -> &MultiscreenConfig

Returns the active configuration.

pub fn train(&mut self, input: TrainInput) -> Result<TrainReport>

Trains the lightweight transition state from token IDs.

pub fn infer_tokens(&self, input_ids: &[u32]) -> Result<InferenceOutput>

Runs deterministic token inference.

pub fn save_weights(&self, path: impl AsRef<Path>) -> Result<()>

Saves the engine’s config, learned state, and training report to a JSON weights file.

The file embeds the full MultiscreenConfig so that load_weights can verify a match before restoring state.

pub fn load_weights(&mut self, path: impl AsRef<Path>) -> Result<TrainReport>

Loads weights from a JSON file into this engine.

Config validation: the config stored in the weights file is compared against this engine’s active config. If they do not match exactly, the load is rejected and a Error::WeightsConfigMismatch is returned. This prevents subtle bugs caused by running a state trained under one configuration through an engine configured differently.

Trait Implementations

impl Clone for MultiscreenEngine

fn clone(&self) -> MultiscreenEngine

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for MultiscreenEngine

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.