Struct FastEmbedder 

pub struct FastEmbedder { /* private fields */ }

Implementations

impl FastEmbedder

pub fn recognized(model: &str) -> bool

Whether model is a recognized fastembed model id (i.e. try_for would return Some), with no side effects — no model download.

pub fn try_for(model: &str) -> Result<Option<Self>>

Some if model is a recognized fastembed model id, else None so the caller can fall back to the bag-of-words embedder.

Trait Implementations

impl Embedder for FastEmbedder

fn id(&self) -> String

Stable id used as the index’s model tag (changing it forces reindex).

fn embed(&self, texts: &[String], kind: EmbedKind) -> Result<Vec<Vec<f32>>>

fn min_similarity(&self) -> f32

Score floor under which the top match is treated as “nothing relevant” and the hook injects nothing. Embedder-specific because cosine distributions differ sharply: the hashed bag-of-words space is sparse, so unrelated text scores near 0 and a low floor works; bge is anisotropic, so even unrelated text cosines ~0.5 and the floor must sit well above that. The default is calibrated for bag-of-words; embedders override it.

fn score_margin(&self) -> f32

Max score gap below the single best match for a co-relevant peer to still be injected. Tighter spaces (bge) need a smaller margin. Default is for bag-of-words.