Struct EmbeddingEngine 

pub struct EmbeddingEngine { /* private fields */ }

The main embedding engine for generating vector embeddings.

This struct is thread-safe and can be shared across async tasks. Internal fields use Arc so CPU-heavy inference can be offloaded to blocking threads via tokio::task::spawn_blocking.

Implementations

impl EmbeddingEngine

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

Create a new embedding engine with the given configuration.

This will download the model from HuggingFace Hub if not cached.

pub fn dimension(&self) -> usize

Get the embedding dimension for the loaded model.

pub fn model(&self) -> EmbeddingModel

Get the model being used.

pub fn device(&self) -> &Device

Get the device being used.

pub async fn embed_query(&self, text: &str) -> Result<Vec<f32>>

Embed a single query text.

For models like E5, this automatically applies the query prefix.

pub async fn embed_queries(&self, texts: &[String]) -> Result<Vec<Vec<f32>>>

Embed multiple query texts.

For models like E5, this automatically applies the query prefix.

pub async fn embed_document(&self, text: &str) -> Result<Vec<f32>>

Embed a single document/passage.

For models like E5, this automatically applies the document prefix.

pub async fn embed_documents(&self, texts: &[String]) -> Result<Vec<Vec<f32>>>

Embed multiple documents/passages.

For models like E5, this automatically applies the document prefix.

pub async fn embed_raw(&self, texts: &[String]) -> Result<Vec<Vec<f32>>>

Embed texts without any prefix (raw embedding).

pub fn estimate_time_ms(&self, text_count: usize, avg_text_len: usize) -> f64

Estimate the time to embed a batch of texts (in milliseconds).

Trait Implementations

impl Debug for EmbeddingEngine

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

Formats the value using the given formatter.