FairKV: Balancing Per-Head KV Cache for Fast Multi-GPU Inference

Re-assign community