No, it is about compressing the KV cache; see How TurboQuant works.