Crate rank_refine 

Rerank search candidates with embeddings. SIMD-accelerated.

You bring embeddings, this crate scores them. No model weights, no inference.

Quick Start

use rank_refine::simd;

// Dense scoring
let score = simd::cosine(&[1.0, 0.0], &[0.707, 0.707]);

// Late interaction (`ColBERT` `MaxSim`)
let query = vec![vec![1.0, 0.0], vec![0.0, 1.0]];
let doc = vec![vec![0.9, 0.1], vec![0.1, 0.9]];
let score = simd::maxsim_vecs(&query, &doc);

Modules

Module	Purpose
simd	SIMD vector ops (dot, cosine, maxsim)
colbert	Late interaction (MaxSim), token pooling
diversity	MMR + DPP diversity selection
crossencoder	Cross-encoder trait
matryoshka	MRL tail refinement

Advanced: scoring for trait-based polymorphism, embedding for type-safe wrappers.

Modules

colbert

Late interaction scoring via MaxSim.

crossencoder

Cross-encoder reranking: f(q, d) = NN([q; d]).

diversity

Diversity-aware reranking algorithms.

embedding

Type-safe embedding wrappers.

matryoshka

Two-stage retrieval using Matryoshka embeddings.

prelude

Common imports for reranking.

scoring

Scoring traits and utilities.

simd

Vector operations with SIMD acceleration.

Structs

RefineConfig

Configuration for score blending and truncation.

Enums

RefineError

Errors from refinement operations.

Functions

as_slices

Convert &[Vec<f32>] to Vec<&[f32]>.

Type Aliases

Result

Result type for refinement operations.