Struct FastembedEmbedder 

pub struct FastembedEmbedder { /* private fields */ }

Implementations

impl FastembedEmbedder

pub fn new(cfg: FastembedEmbedderConfig) -> Result<Self>

Constructor that fetches model files from HuggingFace at runtime. Requires the embedder-hub Cargo feature (pulls hf-hub + native-tls). Embedded library consumers that load model bytes from their own storage should use FastembedEmbedder::from_user_defined_files instead — it works under embedder-core alone.

pub fn from_user_defined_files( cfg: FastembedEmbedderConfig, onnx: Vec<u8>, tokenizer: Vec<u8>, tokenizer_config: Vec<u8>, model_config: Vec<u8>, ) -> Result<Self>

Bytes-in constructor: caller supplies the model files directly. No HuggingFace fetch, no hf-hub dep. Available under embedder-core.

onnx is the raw ONNX model bytes. tokenizer is the raw tokenizer.json bytes. tokenizer_config and model_config carry the JSON bytes from tokenizer_config.json and config.json (used to pin pad-token / max-length tokenizer settings — same shape the HF-fetch path uses).

Pooling is read from cfg.pooling ("cls" or "mean"). cfg.threads controls ORT intra-threads (default 1). cfg.is_byo() is not required — this constructor never consults cfg.hf_repo / cfg.onnx_path.

Used by embedded library consumers (e.g. AIDB pgrx extension) that load model bytes from extension-managed storage rather than fetching at runtime.

pub fn embed_seconds(&self) -> f64

Cumulative wall time spent in embed() calls so far.

pub fn dim(&self) -> usize

pub fn embed(&mut self, texts: Vec<String>) -> Result<Vec<Vec<f32>>>

Embed a batch of texts. Returns a flat Vec<Vec<f32>> ordered to match the input. Verifies the output dim matches the config dim. Tracks cumulative wall time in self.embed_seconds.

Trait Implementations

impl BoundaryEmbedder for FastembedEmbedder

Available on crate feature chunkers only.

Default in-tree implementation of crate::chunker::BoundaryEmbedder for FastembedEmbedder. Routes through FastembedEmbedder::embed (which already takes Vec<String> and returns Vec<Vec<f32>>) by allocating owned Strings from the input slice. Owns one allocation per call; negligible vs. the ORT inference cost.

Present when chunkers is enabled (this module is itself embedder-core- gated, so FastembedEmbedder is always available here). The trait lives in crate::chunker and is always available.

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

Embed a batch of texts. Returns one vector per input, in input order. Implementations must produce vectors of consistent dimension across the batch — SemanticChunker only validates that successive calls return the same shape.