August 5, 2026 SmartGen: Seamless Disaggregated LLM Inference with Selective KV Cache Transfer SmartGen 解决自托管 LLM 推理中 P/D 分离架构的网络瓶颈问题，通过三种 KV 缓存传输路径（Profile-based Proactive、Parallel On-demand、Speculative Transfer）实现无缝分离推理 llm-inference p-d-disaggregation kv-cache network-optimization selective-transfer self-hosted cloud-gpu rdma
July 23, 2026 HyMCache A KV Cache Framework for Multi-Turn LLM Serving with CXL-Hybrid Memory llm-serving kv-cache cxl-memory memory-tiering remote-caching pd-disaggregation rdma