Enum VectorData 

pub enum VectorData {
    Dense(Vec<f32>),
    Sparse(Vec<(u32, f32)>),
    Multi(Vec<Vec<f32>>),
}

Vector data supporting different embedding types.

ManifoldDB supports three types of vector embeddings:

• Dense: Standard floating-point vectors (e.g., BERT, BGE, Jina)
• Sparse: Index-value pairs for keyword/BM25-style matching (e.g., SPLADE)
• Multi: Multiple vectors per entity for late interaction (e.g., ColBERT)

Example

use manifoldb_core::VectorData;

let embedding = VectorData::Dense(vec![0.1; 768]);
assert_eq!(embedding.dimension(), Some(768));

let sparse = VectorData::Sparse(vec![(5, 0.9), (100, 0.5)]);
assert_eq!(sparse.dimension(), None); // Sparse vectors have no fixed dimension

Variants

Dense(Vec<f32>)

Dense floating-point vector.

Used for standard embedding models like BERT, BGE, GTE, Jina, etc. The dimension is fixed and determined by the model.

Sparse(Vec<(u32, f32)>)

Sparse vector with (index, value) pairs.

Used for sparse embedding models like SPLADE, or traditional keyword/BM25-style representations. Only non-zero dimensions are stored.

Multi(Vec<Vec<f32>>)

Multi-vector (array of dense vectors).

Used for late-interaction models like ColBERT where each token gets its own embedding vector.

Implementations

impl VectorData

pub fn dimension(&self) -> Option<usize>

Returns the dimension of the vector, if applicable.

• For Dense: returns the length of the vector
• For Sparse: returns None (sparse vectors have no fixed dimension)
• For Multi: returns the dimension of the first sub-vector, or None if empty

pub const fn is_dense(&self) -> bool

Returns true if this is a dense vector.

pub const fn is_sparse(&self) -> bool

Returns true if this is a sparse vector.

pub const fn is_multi(&self) -> bool

Returns true if this is a multi-vector.

pub fn as_dense(&self) -> Option<&[f32]>

Returns the dense vector data, if this is a dense vector.

pub fn as_sparse(&self) -> Option<&[(u32, f32)]>

Returns the sparse vector data, if this is a sparse vector.

pub fn as_multi(&self) -> Option<&[Vec<f32>]>

Returns the multi-vector data, if this is a multi-vector.

Trait Implementations

impl Clone for VectorData

fn clone(&self) -> VectorData

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for VectorData

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

Formats the value using the given formatter.

impl<'de> Deserialize<'de> for VectorData

fn deserialize<__D>( __deserializer: __D, ) -> Result<VectorData, <__D as Deserializer<'de>>::Error>

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl From<Vec<(u32, f32)>> for VectorData

fn from(v: Vec<(u32, f32)>) -> VectorData

Converts to this type from the input type.

impl From<Vec<Vec<f32>>> for VectorData

fn from(v: Vec<Vec<f32>>) -> VectorData

Converts to this type from the input type.

impl From<Vec<f32>> for VectorData

fn from(v: Vec<f32>) -> VectorData

Converts to this type from the input type.

impl PartialEq for VectorData

fn eq(&self, other: &VectorData) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl Serialize for VectorData

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

where __S: Serializer,

Serialize this value into the given Serde serializer.

impl StructuralPartialEq for VectorData