Struct StaticBackend 

pub struct StaticBackend { /* private fields */ }

Model2Vec static embedding backend.

Holds the vocabulary matrix in memory (optionally memory-mapped for cold-start speed). The tokenizer used is the same tokenizer.json as the source model (BGE-Large / ModernBERT), so token IDs are consistent with the ONNX/Candle paths.

Implementations

impl StaticBackend

pub async fn new(config: &ModelConfig) -> Result<Self>

Build a new StaticBackend.

Downloads vocab_matrix.bin and the model tokenizer from HuggingFace on first run; subsequent calls use the local cache.

pub fn from_matrix( matrix: Vec<f32>, tokenizer: Tokenizer, dimension: usize, ) -> Result<Self>

Build from a pre-loaded vocabulary matrix (useful for tests).

matrix must be a flat row-major array of shape [vocab_size × dimension].

pub fn model2vec_dimension() -> usize

Configured Model2Vec output dimension (DAKERA_MRL_DIM, default 256).

Trait Implementations

impl EmbeddingBackend for StaticBackend

fn embed_batch<'life0, 'life1, 'async_trait>( &'life0 self, texts: &'life1 [String], ) -> Pin<Box<dyn Future<Output = Result<Vec<Vec<f32>>>> + Send + 'async_trait>>

where Self: 'async_trait, 'life0: 'async_trait, 'life1: 'async_trait,

Embed a batch of raw texts and return a float32 embedding per input.

fn dimension(&self) -> usize

Embedding dimensionality (fixed for the lifetime of the backend).

fn backend_kind(&self) -> BackendKind

The concrete backend kind — used for metrics and re-embed bookkeeping.