🧰 Mooncake

The items that name it 4

• FlatCake: A Flat KVCache Storage Architecture for LLM Serving
  paper2026-10-28

  Baseline KV cache system that manages host DRAM and NVMe SSD as a tiered hierarchy.

• Cross-request KV cache exposure in the GLM-5.3 Workers AI model (155 responses)
  writeup2026-10-26

  Transfer engine that moves KV cache from prefill GPUs to decode GPUs over RDMA in Cloudflare's SGLang deployment.

• Computation and Data Movement for Inference
  article2026-09-21

  Orchestration layer for inference clusters, named alongside Dynamo and custom schedulers.

• vLLM v0.30.0
  repo

  KV cache store supporting heterogeneous TP sharing and used as a KV connector

In every workspace you can read 5