Struct BinaryDenseVectorConfig 

pub struct BinaryDenseVectorConfig {
    pub dim: usize,
    pub index_type: BinaryIndexType,
    pub num_clusters: Option<usize>,
    pub target_vectors: Option<u64>,
    pub tree_levels: Option<u8>,
    pub ivf_routing: IvfRoutingMode,
    pub nprobe: usize,
    pub soar: Option<SoarConfig>,
}

Configuration for binary dense vector fields

Binary dense vectors store packed bits (1 bit per dimension) and use Hamming distance for scoring. Segments accumulate exact packed codes and use the same global IVF router after build_vector_index.

Fields

dim: usize

Number of bits (dimensions). Storage is ceil(dim/8) bytes per vector.

index_type: BinaryIndexType

ANN index type: Flat (brute-force SIMD Hamming) or Ivf (default) (k-majority Hamming clusters — probe nprobe clusters at query time). IVF pays off for segments past a few million vectors.

num_clusters: Option<usize>

Number of IVF leaf clusters, selected from corpus and sample size by default.

target_vectors: Option<u64>

Expected steady-state vector count used only for automatic topology sizing. Training readiness still depends on the observed live corpus. Explicit num_clusters takes precedence over this hint.

tree_levels: Option<u8>

Number of levels in the ScaNN Hamming routing tree. When omitted, training derives the depth from corpus size. Only meaningful for ScaNN.

ivf_routing: IvfRoutingMode

Coarse-codebook routing strategy. Uses the same routing planner as floating-point IVF indexes.

nprobe: usize

Clusters to probe during search (default: 64)

soar: Option<SoarConfig>

Optional one-secondary selective spilling for binary ScaNN. The alternate leaf is chosen by exact centroid Hamming distance. Unlike float SOAR, packed bits have no continuous residual geometry.

Implementations

impl BinaryDenseVectorConfig

pub fn new(dim: usize) -> Self

pub fn with_ivf(self, num_clusters: Option<usize>, nprobe: usize) -> Self

Enable the IVF index (builder pattern)

pub fn with_target_vectors(self, target_vectors: u64) -> Self

Hint the expected steady-state corpus size for automatic topology.

pub fn with_ivf_routing(self, routing: IvfRoutingMode) -> Self

Set flat, two-level, or HNSW IVF centroid routing explicitly.

pub fn with_soar(self, soar: SoarConfig) -> Self

Enable selective secondary-leaf spilling for binary ScaNN.

pub fn without_soar(self) -> Self

Disable binary ScaNN secondary-leaf spilling.

pub fn optimal_num_clusters(&self, num_vectors: usize) -> usize

Balanced binary IVF cluster count for a given vector count.

pub fn byte_len(&self) -> usize

Number of bytes needed to store one vector

Trait Implementations

impl Clone for BinaryDenseVectorConfig

fn clone(&self) -> BinaryDenseVectorConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for BinaryDenseVectorConfig

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for BinaryDenseVectorConfig

fn deserialize<__D>(__deserializer: __D) -> Result<Self, __D::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl Serialize for BinaryDenseVectorConfig

fn serialize<__S>(&self, __serializer: __S) -> Result<__S::Ok, __S::Error>

where __S: Serializer,

Serialize this value into the given Serde serializer.