Show HN: TurboQuant for vector search – 2-4 bit compression
Summary
TurboQuant for vector search compresses high-dimensional vectors to 2-4 bits per coordinate with zero training, enabling fast, data-oblivious similarity search. The project is a Rust-based implementation with Python bindings (via PyO3), and it includes architecture details, SIMD-accelerated kernels, and benchmark comparisons against FAISS across CPU architectures.