Back to browse
GitHub Repository

A vector index built on TurboQuant, written in Rust with Python bindings

3,756 starsPython

TurboQuant for vector search – 2-4 bit compression

by justsomeguy1996·Mar 29, 2026·89 points·6 comments

AI Analysis

●●SolidBig BrainNiche Gem

Data-oblivious quantization beats Product Quantization on online updates.

Strengths
  • Random rotation + Lloyd-Max quantization achieves 16x compression with near-optimal distortion
  • No training required means vectors can be added without rebuilding the index
  • Reproduces paper benchmarks with actual recall@k measurements on M3 Max
Weaknesses
  • Unofficial implementation competing against FAISS, Milvus, and Pinecone's built-in compression
  • 97 stars suggests limited real-world adoption or production testing so far
Category
Target Audience

ML engineers building vector search systems

Similar To

FAISS · SPTAG · DiskANN

Similar Projects

AI/ML●●●Banger

TurboQuant-WASM – Google's vector quantization in the browser

Google's ICLR 2026 quantization paper running client-side with SIMD-accelerated dot products.

WizardryZero to One
teamchong
16572mo ago
AI/MLMid

Standalone TurboQuant KV Cache Inference

Standalone KV cache compression script implementing TurboQuant with 1.55x ratio.

Big BrainShip It
g023
342mo ago
AI/ML●●●Banger

UltraCompress – first mathematically lossless 5-bit LLM compression

Runs 405B model compression on a single 32GB GPU when others need enterprise clusters.

WizardryBig Brain
mounnar
6026d ago