Turbo Quant

https://turbo-quant.com
It uses a two stage approach PolarQuant for polar coordinate rotation and scalar quantization followed by a 1 bit QJL residual correction to achieve near lossless compression at

https://github.com › turboquant
Implementation of TurboQuant KV cache compression ICLR 2026 arXiv 2504 19874 with vLLM integration Tested on dense and MoE architectures across RTX 3090 and RTX 5090 GPUs Setup

https://en.wikipedia.org › wiki › TurboQuant
TurboQuant consists of two related algorithms TurboQuantmse which is optimized for mean squared error MSE and TurboQuantprod which is optimized for unbiased inner product estimation

https://www.baseten.co › blog
Our method transforms the KV embeddings into polar coordinates using a recursive algorithm and then quantizes resulting angles The long context evaluation demonstrates that

https://dev.to › arshtechpro › turboquant-what...
Published by Google Research on March 24 2026 and headed to ICLR 2026 TurboQuant is a compression algorithm that shrinks the KV cache the biggest memory bottleneck

https://arxiv.org › abs
We propose TurboQuant to address both mean squared error MSE and inner product distortion overcoming limitations of existing methods that fail to achieve optimal distortion rates

https://www.tomshardware.com › tech-industry › ...
Google Research published TurboQuant on Tuesday a training free compression algorithm that quantizes LLM KV caches down to 3 bits without any loss in model accuracy

https://www.marktechpost.com › google-introduces-turboquant
Google research team has proposed TurboQuant a data oblivious quantization framework designed to achieve near optimal distortion rates for high dimensional Euclidean vectors

https://www.linkedin.com › pulse › turboquant-new-way...
Normally as a conversation gets longer this cache balloons until the system runs out of VRAM Video RAM TurboQuant from Google uses a two stage mathematical process to compress