Struct TextEmbedding 

pub struct TextEmbedding<B: Backend = DefaultBackend> { /* private fields */ }

Minimal text embedding interface inspired by fastembed-rs.

Implementations

impl TextEmbedding<DefaultBackend>

pub async fn new(options: TextEmbeddingOptions) -> Result<Self>

Loads a MiniLM text embedding model onto the default WGPU device.

impl<B> TextEmbedding<B>

where B: Backend,

pub async fn new_with_device( device: &B::Device, options: TextEmbeddingOptions, ) -> Result<Self>

Loads a MiniLM text embedding model onto the provided device.

pub fn embed(&self, document: impl AsRef<str>) -> Result<Vec<f32>>

Embeds a single document and returns one embedding vector.

pub fn embed_query(&self, query: impl AsRef<str>) -> Result<Vec<f32>>

Embeds a search query using any model-specific retrieval prompt.

Some retrieval models train queries and documents with different text prefixes. Use this when the text is the thing being searched for. Use TextEmbedding::embed when the text is the content being indexed.

pub fn embed_batch<S: AsRef<str>>( &self, documents: &[S], batch_size: Option<usize>, ) -> Result<Vec<Vec<f32>>>

Embeds documents in batches and returns one vector per input string.

pub fn embed_query_batch<S: AsRef<str>>( &self, queries: &[S], batch_size: Option<usize>, ) -> Result<Vec<Vec<f32>>>

Embeds search queries in batches using model-specific retrieval prompts.

See TextEmbedding::embed_query for when query embeddings differ from document embeddings.

pub fn model(&self) -> EmbeddingModel

Returns the loaded embedding checkpoint.

Trait Implementations

impl<B: Debug + Backend> Debug for TextEmbedding<B>

where B::Device: Debug,

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

Formats the value using the given formatter.