Struct HealingDecoder 

pub struct HealingDecoder {
    pub healer: TokenHealer,
}

Combines token healing with a simple token-by-token generation loop.

Healing is applied once to the prompt; then max_tokens additional tokens are drawn using the sample closure.

Fields

healer: TokenHealer

The inner healer driving the healing step.

Implementations

impl HealingDecoder

pub fn new(config: TokenHealingConfig) -> Self

Create a new decoder with the supplied healing configuration.

pub fn generate<F, G>( &self, prompt_tokens: Vec<u32>, vocab_size: usize, max_tokens: usize, get_logits: F, sample: G, ) -> (HealingResult, Vec<u32>)

where F: FnMut(&[u32]) -> Vec<f32>, G: FnMut(Vec<f32>) -> u32,

Apply token healing to prompt_tokens, then generate up to max_tokens additional tokens.

Parameters

• get_logits — called with the current token sequence; returns logits.
• sample — called with the raw logits; returns the next token id.

Returns

A pair (HealingResult, generated_tokens).