turboquant-pytorch

PyTorch implementation of TurboQuant for LLM KV cache compression using two-stage vector quantization (random rotation + Lloyd-Max + QJL residual correction).

Independent GitHub snapshot · Automated classification · Original documentation governs