Struct QuantizedEmbeddingCache 

pub struct QuantizedEmbeddingCache { /* private fields */ }

Quantized embedding cache with scalar or product quantization and asymmetric distance computation for compressed similarity search.

Implementations

impl QuantizedEmbeddingCache

pub fn new(config: QuantizedCacheConfig, dimensions: usize) -> Self

Create a new, untrained cache.

pub fn train(&mut self, training_vectors: &[Vec<f32>]) -> Result<()>

Train quantization parameters from training_vectors.

Must be called before any calls to add or search.

pub fn add(&mut self, id: String, vector: Vec<f32>) -> Result<()>

Compress and store a vector by id.

pub fn add_with_metadata( &mut self, id: String, vector: Vec<f32>, metadata: HashMap<String, String>, ) -> Result<()>

Compress and store a vector with metadata.

pub fn get(&self, id: &str) -> Option<Vec<f32>>

Retrieve the decompressed (reconstructed) vector for id.

pub fn len(&self) -> usize

Number of vectors in the cache.

pub fn is_empty(&self) -> bool

Returns true if no vectors are stored.

pub fn search(&mut self, query: &[f32], k: usize) -> Result<Vec<(String, f32)>>

Find the k nearest cached vectors to query using asymmetric distance.

The query is kept in fp32; each database code is decompressed on-the-fly and Euclidean distance is computed.

pub fn metrics(&self) -> &CacheMetrics

Return a snapshot of current metrics.

pub fn config(&self) -> &QuantizedCacheConfig

Access the configuration.