TurboQuant is a vector-embedding compression technique that applies a random orthogonal rotation to embedding coordinates, then maps the rotated values onto a fixed Lloyd-Max codebook and stores compact indices instead of floating-point values. It supports 1, 1.5, 2, and 4-bit depths, yielding compression ratios from 8x to 32x, and Qdrant’s implementation adds length renormalization and per-coordinate calibration to reduce scoring error. Across three information-retrieval benchmarks, TurboQuant outperforms binary quantization at matching compression levels and matches scalar quantization quality at half the code size.
