Struct ConfigurableSampler 

pub struct ConfigurableSampler {
    pub config: SamplingConfig,
    /* private fields */
}

A sampler that combines temperature scaling, top-k filtering, top-p (nucleus) filtering, and repetition penalty into a single configurable pipeline.

Pipeline order: repetition_penalty → temperature → top-k → top-p → sample.

Fields

config: SamplingConfig

The complete sampling configuration.

Implementations

impl ConfigurableSampler

pub fn new(config: SamplingConfig) -> Result<Self, SamplingError>

Construct a ConfigurableSampler from a SamplingConfig.

Validates temperature, top_k, and top_p at construction time.

pub fn with_default() -> Self

Construct a ConfigurableSampler with the default configuration.

This is equivalent to ConfigurableSampler::new(SamplingConfig::default()) but is infallible because the defaults are always valid.

pub fn apply_repetition_penalty( logits: &mut [f64], context: &[usize], penalty: f64, )

Apply repetition penalty in-place.

For each token that appears in context:

• If the logit is positive → divide by penalty (move toward 0).
• If the logit is negative → multiply by penalty (move away from 0).

A penalty of 1.0 is a no-op.

pub fn sample( &mut self, logits: &[f64], context: &[usize], ) -> Result<SampledToken, SamplingError>

Run the full sampling pipeline:

1. Apply repetition penalty to logits for tokens in context.
2. Scale by temperature.
3. Apply top-k filter (if configured).
4. Apply top-p (nucleus) filter (if configured).
5. Sample from the resulting distribution.

Trait Implementations

impl Debug for ConfigurableSampler

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

Formats the value using the given formatter.