Struct Normalizer 

pub struct Normalizer {
    model: LlamaModel,
}

A loaded normalizer. Holds the weights; cheap to call repeatedly.

Fields

model: LlamaModel

Implementations

impl Normalizer

pub fn load(path: &Path) -> Result<Self, NormalizeError>

Load weights from a GGUF file.

The architecture is checked against CAUSAL_ARCHS BEFORE the load - see NormalizeError::NotCausal for why that order is the whole point of the check.

pub fn normalize(&self, text: &str) -> Result<String, NormalizeError>

Rewrite one raw transcript as written text, at the shipped defaults.

An empty result is a valid answer, not a failure: input that is nothing but filler normalizes to nothing.

pub fn normalize_with( &self, text: &str, opts: NormalizeOptions, ) -> Result<String, NormalizeError>

As Normalizer::normalize, with the register and decoder settings chosen per call. Nothing here is cached against them, so they cost a string format and can change on every call.

fn run( &self, prompt: &str, source: &str, opts: NormalizeOptions, ) -> Result<String, NormalizeError>

source is what speculation drafts from, and is the transcript rather than the prompt wrapped around it.

fn session(&self, prompt: &str, k: usize) -> Result<Session<'_>, NormalizeError>

Decode the prompt and hand back everything generation needs.

Trait Implementations

impl Debug for Normalizer

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

Formats the value using the given formatter.