Run LMCache MP to share KV caches among vLLM pods in Kubernetes clusters.
npx skills add https://github.com/air-gapped/skills --skill lmcache-mp
Please help me install this Agent Skill. Skill: lmcache-mp Source: https://github.com/air-gapped/skills/tree/main/.claude/skills/lmcache-mp Command: npx skills add https://github.com/air-gapped/skills --skill lmcache-mp
LMCache multiprocess (MP) mode runs the KV-cache server in its own process so multiple vLLM pods on a node can share a centralized cache with independent CPU/memory budgets and a ZMQ-based transport, enabling scalable inference at scale.
Start a LMCache MP DaemonSet on your cluster and configure vLLM to connect to the LMCache MP port.
High-intent search queries and answers about installing and using this skill.
Sharing KV caches across vLLM pods requires a separate LMCache multiprocess server per node deployed as a DaemonSet, serving all pods via ZMQ transport.
Offloading vLLM KV cache to a dedicated LMCache multiprocess server reduces GPU contention by isolating cache management in its own process with independent CPU and memory budgets separate from the inference pods.
Configure L1 and L2 cache storage cascades for vLLM pods by selecting adapters like nixl_store, fs, mooncake_store, or s3 within your Kubernetes Deployment manifests.
Yes, LMCache multiprocess mode requires hostNetwork configuration and /dev/shm sharing for inter-process communication between the DaemonSet and vLLM pods.
No, LMCache MP requires aligned vLLM and LMCache versions plus correct startup flags to ensure proper connectivity between the cache server and inference pods.
Connection failures between vLLM and LMCache MP typically stem from misaligned versions, incorrect startup flags, or missing hostNetwork and /dev/shm configurations for IPC.
ovachieverDeploy OpenAI-compatible LLM APIs with vLLM for high-throughput inference.
air-gappedCoordinate vLLM KV caching across native offload, LMCache, Nixl, and Mooncake backends.
beyondkmpTranslate an algorithmic philosophy into a self-contained P5.js HTML artwork.
Ikram-AlamAutomate PostgreSQL cluster deployment on Kubernetes with CloudNativePG.
