KV Cache Quantization

TurboQuant KV Cache Quantization Beats Baselines on Gemma 4 and Qwen

Community benchmarks show TurboQuant KV quantization achieves near-zero accuracy loss at 3.1 bits on Gemma 4 with 34% long-context speedup.