Crate rig_fastembed 

Local embedding model integration backed by fastembed.

A loaded fastembed model is the Fastembed transport, the runtime behind a local embedding wire (text_embeddings) that embeds in the calling process. The default feature set enables Hugging Face model downloads and ONNX Runtime binary downloads.

use rig_core::Model;
use rig_fastembed::{Fastembed, FastembedModel, text_embeddings};

let model = Fastembed::load(&FastembedModel::AllMiniLML6V2Q)?.embedding(&FastembedModel::AllMiniLML6V2Q, None)?;

rig-fastembed is native-only and does not target wasm32-unknown-unknown. The root rig facade re-exports this crate as rig::fastembed when one of its Fastembed features is enabled.

Structs

Fastembed

A loaded Fastembed model: the transport that embeds in the calling process. Clones share the loaded model. Pair it with the text_embeddings wire of the model it loaded: the wire names the model and width that spans and capabilities report, and the transport embeds with whatever it loaded. In-process execution reports no raw payload, usage, or request id.

Enums

FastembedError

Errors raised while resolving or initializing a Fastembed model.

FastembedModel

Functions

text_embeddings

The local embedding wire of model at ndims dimensions, named fastembed and addressing the model by its fastembed name. None takes the width from the model metadata, which errors for models fastembed does not know.