KVarN: Native vLLM backend for KV-cache quantization by Huawei

KVarN is a native vLLM KV-cache quantization backend for your agents: 3-5x more context, throughput above FP16, and FP16-level accuracy. Calibration-free, one flag. - huawei-csl/KVarN