Struct EmbedderRegistry 

pub struct EmbedderRegistry { /* private fields */ }

Thread-safe registry that holds the active embedding backends.

On creation the TF-IDF slot is empty and the IdentityEmbedder is used as a deterministic fall-back. Once enough documents have been seen (or EmbedderRegistry::fit_tfidf is called explicitly) the TF-IDF embedder is installed and used for all subsequent requests.

Implementations

impl EmbedderRegistry

pub fn new(default_dim: usize) -> Self

Create a new registry.

default_dim controls the dimensionality of the IdentityEmbedder fallback and is also used as max_features when fitting TF-IDF.

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

Embed a slice of text strings, returning one dense vector per input.

Uses the TF-IDF backend when it has been fitted; falls back to IdentityEmbedder otherwise. Texts that fail to embed are silently replaced with a zero vector of the appropriate dimension.

pub fn fit_tfidf(&self, corpus: &[String])

Fit the TF-IDF backend from corpus.

After this call embed_texts will use TF-IDF for all subsequent requests. Subsequent calls replace the existing model.

pub fn embedding_dim(&self) -> usize

Return the current embedding dimension.

Returns the TF-IDF vocabulary size when a fitted model is present, otherwise the configured default_dim.

pub fn encode_base64(embedding: &[f32]) -> String

Encode an embedding vector as a hex string (pure Rust, no external deps).

Each f32 is serialised as four bytes in little-endian order, with each byte represented as two lowercase hex digits. The result is therefore 8 * embedding.len() characters long.