Struct SparseQueryConfig 

pub struct SparseQueryConfig {
    pub tokenizer: Option<String>,
    pub weighting: QueryWeighting,
    pub heap_factor: f32,
    pub weight_threshold: f32,
    pub max_query_dims: Option<usize>,
    pub pruning: Option<f32>,
}

Query-time configuration for sparse vectors

Research-validated query optimization strategies:

• weight_threshold (0.01-0.05): Drop query dimensions with weight below threshold
  • Filters low-IDF tokens that add latency without improving relevance
• max_query_dims (10-20): Process only top-k dimensions by weight
  • 30-50% latency reduction with <2% nDCG loss (Qiao et al., 2023)
• heap_factor (0.8): Skip blocks with low max score contribution
  • ~20% speedup with minor recall loss (SEISMIC-style)

Fields

tokenizer: Option<String>

HuggingFace tokenizer path/name for query-time tokenization Example: “Alibaba-NLP/gte-Qwen2-1.5B-instruct”

weighting: QueryWeighting

Weighting strategy for tokenized query terms

heap_factor: f32

Heap factor for approximate search (SEISMIC-style optimization) A block is skipped if its max possible score < heap_factor * threshold

Research recommendation:

• 1.0 = exact search (default)
• 0.8 = approximate, ~20% faster with minor recall loss (RECOMMENDED for production)
• 0.5 = very approximate, much faster but higher recall loss

weight_threshold: f32

Minimum weight for query dimensions (query-time pruning) Dimensions with abs(weight) below this threshold are dropped before search. Useful for filtering low-IDF tokens that add latency without improving relevance.

• 0.0 = no filtering (default)
• 0.01-0.05 = recommended for SPLADE/learned sparse models

max_query_dims: Option<usize>

Maximum number of query dimensions to process (query pruning) Processes only the top-k dimensions by weight

Research recommendation (Multiple papers 2022-2024):

• None = process all dimensions (default, exact)
• Some(10-20) = process top 10-20 dimensions only (RECOMMENDED for SPLADE)
  • 30-50% latency reduction
  • <2% nDCG@10 loss

pruning: Option<f32>

Fraction of query dimensions to keep (0.0-1.0), same semantics as indexing-time pruning: sort by abs(weight) descending, keep top fraction. None or 1.0 = no pruning.

Trait Implementations

impl Clone for SparseQueryConfig

fn clone(&self) -> SparseQueryConfig

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for SparseQueryConfig

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

Formats the value using the given formatter.

impl Default for SparseQueryConfig

fn default() -> Self

Returns the “default value” for a type.

impl<'de> Deserialize<'de> for SparseQueryConfig

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

where __D: Deserializer<'de>,

Deserialize this value from the given Serde deserializer.

impl PartialEq for SparseQueryConfig

fn eq(&self, other: &SparseQueryConfig) -> 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 SparseQueryConfig

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

where __S: Serializer,

Serialize this value into the given Serde serializer.

impl StructuralPartialEq for SparseQueryConfig