Struct RerankerConfig 

pub struct RerankerConfig {
    pub field: Field,
    pub vector: Vec<f32>,
    pub binary_vector: Vec<u8>,
    pub combiner: MultiValueCombiner,
    pub unit_norm: bool,
    pub matryoshka_dims: Option<usize>,
    pub rrf_k: f32,
}

Configuration for L2 dense/binary vector reranking

Fields

field: Field

Vector field (dense or binary dense)

vector: Vec<f32>

Query vector (f32, for dense fields)

binary_vector: Vec<u8>

Query vector (packed bits, for binary dense fields). When non-empty, Hamming distance scoring is used instead of cosine.

combiner: MultiValueCombiner

How to combine scores for multi-valued documents

unit_norm: bool

Whether stored vectors are pre-normalized to unit L2 norm. When true, scoring uses dot-product only (skips per-vector norm — ~40% faster). Ignored for binary fields.

matryoshka_dims: Option<usize>

Matryoshka pre-filter: number of leading dimensions to use for cheap approximate scoring before full-dimension exact reranking. Ignored for binary fields.

rrf_k: f32

Reciprocal Rank Fusion k parameter. When > 0, fuses L1 (first-stage) and L2 (reranker) rankings: score(d) = 1/(k + rank_L1) + 1/(k + rank_L2). Typical value: 60. When 0, RRF is disabled and only L2 scores are used.

Trait Implementations

impl Clone for RerankerConfig

fn clone(&self) -> RerankerConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for RerankerConfig

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

Formats the value using the given formatter.