Struct FixedDistDraftModel 

pub struct FixedDistDraftModel { /* private fields */ }

Draft model that always returns the same categorical distribution and samples tokens from it.

Implementations

impl FixedDistDraftModel

pub fn new(probs: Vec<f64>) -> SpeculativeDecodingResult<Self>

Build a mock draft from a linear-space probability vector. The vector must be non-empty and sum to ≈ 1.

pub fn probs(&self) -> &[f64]

Access the (normalized) linear-space probabilities.

pub fn logprobs(&self) -> &[LogProb]

Access the log-probability row.

Trait Implementations

impl Clone for FixedDistDraftModel

fn clone(&self) -> FixedDistDraftModel

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for FixedDistDraftModel

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

Formats the value using the given formatter.

impl DraftModel for FixedDistDraftModel

fn vocab_size(&self) -> usize

Vocabulary cardinality the model emits log-probs over.

fn propose( &self, _prefix: &[TokenId], k: usize, rng: &mut dyn SpecRng, ) -> SpeculativeDecodingResult<DraftProposal>

Extend prefix by k draft tokens; return the chosen tokens, their log-probabilities and the full per-position distributions.