Struct FastEmbedEmbedder 

pub struct FastEmbedEmbedder {
    model_name: &'static str,
}

FastEmbed embedder using all-MiniLM-L6-v2.

Uses the fastembed-rs library for real semantic embeddings. The model is lazily loaded on first embed call to preserve cold start time.

Fields

model_name: &'static str

Model name for logging/debugging.

Implementations

impl FastEmbedEmbedder

pub const DEFAULT_DIMENSIONS: usize = DEFAULT_DIMENSIONS

Default embedding dimensions for all-MiniLM-L6-v2.

pub const fn new() -> Self

Creates a new FastEmbed embedder.

Note: Model is lazily loaded on first embed() call.

pub const fn with_dimensions(_dimensions: usize) -> Self

Creates a new embedder with custom dimensions.

Note: This is provided for API compatibility but dimensions are fixed by the model (384 for all-MiniLM-L6-v2).

fn get_model() -> Result<&'static Mutex<TextEmbedding>>

Gets or initializes the embedding model (thread-safe).

Performance Note

The model is loaded lazily on first use to preserve cold start time. Subsequent calls return the cached instance.

The first call blocks synchronously (~100-500ms) while loading the ONNX model. This is an intentional design decision:

• One-time cost amortized over all subsequent calls (instant)
• Sync API is simpler and doesn’t require async runtime everywhere
• Alternative (tokio::spawn_blocking) would require async Embedder trait

For applications sensitive to first-call latency, consider warming up the embedder during startup: FastEmbedEmbedder::new().embed("warmup").ok();

pub const fn model_name(&self) -> &'static str

Returns the model name.

Trait Implementations

impl Default for FastEmbedEmbedder

fn default() -> Self

Returns the “default value” for a type.

impl Embedder for FastEmbedEmbedder

fn dimensions(&self) -> usize

Returns the embedding dimensions.

fn embed(&self, text: &str) -> Result<Vec<f32>>

Generates an embedding for the given text.

fn embed_batch(&self, texts: &[&str]) -> Result<Vec<Vec<f32>>>

Generates embeddings for multiple texts.