Struct FastEmbedder 

pub struct FastEmbedder { /* private fields */ }

Local CPU embedder backed by fastembed-rs (ONNX runtime, all-MiniLM-L6-v2).

Why: Default to local-only embeddings so consumers have zero external network dependency and predictable latency. The LRU cache keeps the hot path free of redundant ONNX work for repeat strings (queries, common chunks). What: wraps a single TextEmbedding behind a parking_lot::Mutex (the underlying embed requires &mut self) and an LruCache<String, Vec<f32>>. Initialisation warms the ORT graph with a small batch so the first user query doesn’t pay the one-shot compile cost. Test: embed_batch_returns_correct_dim and cache_hit_is_idempotent (marked #[ignore] — they download a real model).

Implementations

impl FastEmbedder

pub async fn new() -> Result<Self>

Construct a new FastEmbedder with the default cache size.

pub async fn with_cache_size(capacity: usize) -> Result<Self>

Construct with an explicit LRU capacity.

Trait Implementations

impl Embedder for FastEmbedder

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 texts. Returns one Vec<f32> per input, each of length self.dimension(). An empty input batch returns an empty Vec.

fn dimension(&self) -> usize

Output dimension of the produced embeddings.