KVarN: The Calibration-Free KV-Cache Quantization That Keeps Both Speed and Accuracy

The KV-cache quantization story has been a choice between losing speed and losing accuracy. A seven-day-old vLLM backend from Huawei CSL just changed the terms.