Struct Model2VecBackend 

pub struct Model2VecBackend { /* private fields */ }

Static-embedding backend powered by model2vec-rs.

Construct via Self::load — the call is async and offloads the (blocking) StaticModel::from_pretrained download/load onto tokio::task::spawn_blocking.

Implementations

impl Model2VecBackend

pub async fn load(model_type: EmbeddingModelType) -> Result<Self>

Download (if needed) and load a Model2Vec model from the HuggingFace Hub. Verifies the runtime embedding dimension against the EmbeddingModelType constant and returns an error on mismatch — silent drift would corrupt every downstream HNSW index.

Trait Implementations

impl EmbeddingBackend for Model2VecBackend

fn is_bert_based(&self) -> bool

Model2Vec is not BERT — the engine’s BERT branch (concurrency controller, adaptive batching, timeout) is bypassed. The static path dispatches directly because each call is ms-cheap.

fn embedding_dimension(&self) -> usize

Dimension of the produced embedding vectors.

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

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

Process a batch of texts and return one embedding vector per text. The engine slices large inputs into chunks no larger than the engine’s current_batch_size before calling this method.