Vector Quantization Python Packages

vector-quantize-pytorch

Vector (and Scalar) Quantization, in Pytorch

minisom

:red_circle: MiniSom is a minimalistic implementation of the Self Organizing Maps

annlite

⚡ A fast embedded library for approximate nearest neighbor search

gridbook

Out-of-tree vLLM plugin and open format spec for NVFP4-CB / FP8-CB product-codebook weights — 2-6 bit-per-weight LLM quantization served on native Blackwell tensor cores.

markov-kmeans

Early stopping for K-means: stop when only boundary points still switch clusters, and settle them from their own label history. Shorter runs, uncertainty flags, soft memberships. Exact Lloyd, mini-batch and EMA (constant-step) training. scikit-learn compatible.

medtokenizers

Tokenizers for 2D and 3D medical images: VQ, LFQ, FSQ, Residual FSQ, VAE, and AE quantizers behind one backbone.

pyvq

A vector quantization library for Rust :crab: with Python bindings 🐍

dbgsom

A scikit-learn compatible Python implementation of the Directed Batch Growing Self-Organizing Map

llvq-tetra

Tetra : Leech lattice vector quantization for 2-bit LLM weights, in Rust, with a fused multi-shell CUDA decoder.

helix-substrate

Calibration-free HXQ weight compression. Transformers, SSMs, hybrids, MoEs, vision — same codec, same command.

turboquant-space

SIMD-accelerated 4/8-bit vector quantization for approximate nearest neighbor search, based on TurboQuant (ICLR 2026). Standalone C++17 library with Python bindings

time-causal-vae

Time-causal financial generative models: refactored TC-VAE baselines with causal VQ/RVQ tokenizers, token priors, S&P500/VIX, Hawkes/SVMHJD, multi-dimensional benchmarks, and path-risk diagnostics.

timevqvae

[official] PyTorch implementation of TimeVQVAE from the paper ["Vector Quantized Time Series Generation with a Bidirectional Prior Model", AISTATS 2023]

swarndb

The data layer for AI with Vector, Graph and Vector-Math built into one single rust engine

titok-pytorch

Implementation of TiTok, proposed by Bytedance in "An Image is Worth 32 Tokens for Reconstruction and Generation"