Crate next_plaid 

Next-Plaid: CPU-based PLAID implementation for multi-vector search

This crate provides a pure Rust, CPU-only implementation of the PLAID algorithm for efficient multi-vector search (late interaction retrieval).

Re-exports

pub use codec::ResidualCodec;

pub use delete::delete_from_index;

pub use error::Error;

pub use error::Result;

pub use index::MmapIndex;

pub use index::IndexConfig;

pub use index::Metadata;

pub use kmeans::compute_centroids;

pub use kmeans::compute_centroids_from_documents;

pub use kmeans::compute_kmeans;

pub use kmeans::estimate_num_partitions;

pub use kmeans::ComputeKmeansConfig;

pub use search::QueryResult;

pub use search::SearchParameters;

pub use update::UpdateConfig;

Modules

codec

Residual codec for quantization and decompression

delete

Document deletion functionality for removing documents from an existing index.

embeddings

Embedding reconstruction from compressed index data.

error

Error types for the next-plaid library

filtering

SQLite-based metadata filtering for next-plaid indices.

index

Index creation and management for PLAID

kmeans

K-means clustering integration using fastkmeans-rs.

maxsim

High-performance MaxSim scoring for late-interaction (ColBERT) workflows.

mmap

Memory-mapped file support for efficient large index loading.

search

Search functionality for PLAID

update

Index update functionality for adding new documents.

utils

Utility functions for next-plaid

Structs

FastKMeans

Fast k-means clustering implementation compatible with ndarray.

KMeansConfig

Configuration for the FastKMeans algorithm