embed

package
v0.1.0 Latest Latest
Warning

This package is not in the latest version of its module.

 Go to latest Published: Jun 26, 2026 License: Apache-2.0 Imports: 11 Imported by: 0

Documentation

Overview

Package embed provides an OpenAI-compatible embeddings client and the similarity + fusion primitives for hybrid (BM25 + vector) catalog retrieval.

It is the in-house alternative to a Python embedding sidecar or a vector-DB dependency: embeddings are served by the same OpenAI-compatible endpoint as the model (in-cluster vLLM, Ollama, OpenAI), and similarity is brute-force cosine over the (small) catalog — no new datastore, still a single static binary.

This package is the foundation only; wiring hybrid retrieval into the catalog index and the recall gates is an eval-validated follow-up (the BM25-tuned recall thresholds must be re-measured for fused scores, not guessed).

Index

Constants

This section is empty.

Variables

This section is empty.

Functions

func Cosine 

func Cosine(a, b []float32) float64

Cosine returns the cosine similarity of two equal-length vectors, in [-1, 1]. Mismatched-length or zero-norm vectors return 0 ("no signal").

Types

type Client 

type Client struct {
	// contains filtered or unexported fields
}

Client calls an OpenAI-compatible /embeddings endpoint.

func New 

func New(baseURL, model, apiKey string) *Client

New builds an embeddings client. apiKey may be empty (keyless vLLM/Ollama).

func (*Client) Embed 

func (c *Client) Embed(ctx context.Context, texts []string) ([][]float32, error)

Embed returns one vector per input text, in input order. Empty input → nil.

type Fused 

type Fused struct {
	ID    string
	Score float64
}

Fused is an id with its fused relevance score.

func FuseRRF 

func FuseRRF(k float64, rankings ...[]string) []Fused

FuseRRF combines several rankings of the same id space by Reciprocal Rank Fusion: score(id) = Σ 1/(k + rank), rank 0-based within each ranking; an id absent from a ranking contributes nothing. RRF is scale-free — it fuses BM25 and cosine rankings without their different score magnitudes distorting the result. k dampens low ranks (60 is the common default). Results are sorted by descending fused score, ties broken by id for determinism.

Source Files

View all Source files

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL
go.dev uses cookies from Google to deliver and enhance the quality of its services and to analyze traffic. Learn more.