SwiftCache: Efficient LLM Serving for Multi-turn Conversations with Heterogeneous KV Cache Sharing

Hu, Jianmin, Xu, Minxian, Wang, Sa · arXiv · 2026