lmcache-mp

Official

LMCache MP for scalable KV caching across pods

Authorair-gapped
Version1.0.0
Installs0

System Documentation

What problem does it solve?

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.

Core Features & Use Cases

  • Separate LMCache server per node that serves multiple vLLM pods on the same node, reducing GPU contention and improving cache hit rates for multi-pod deployments.
  • Kubernetes-friendly deployment pattern: a DaemonSet for LMCache and a Deployment for vLLM pods, using L1/L2 storage cascades (nixl_store, nixl_store_dynamic, fs, mooncake_store, s3) as configured.
  • Production-ready references: operator-based reconciliation and example manifests, plus detailed deployment and troubleshooting guidance.

Quick Start

Start a LMCache MP DaemonSet on your cluster and configure vLLM to connect to the LMCache MP port.

Dependency Matrix

Required Modules

None required

Components

references

💻 Claude Code Installation

Recommended: Let Claude install automatically. Simply copy and paste the text below to Claude Code.

Please help me install this Skill:
Name: lmcache-mp
Download link: https://github.com/air-gapped/skills/archive/main.zip#lmcache-mp

Please download this .zip file, extract it, and install it in the .claude/skills/ directory.
View Source Repository

Agent Skills Search Helper

Install a tiny helper to your Agent, search and equip skill from 510,000+ vetted skills library on demand.