vllm-caching

Official

Production KV caching for production-grade vLLM

Authorair-gapped
Version1.0.0
Installs0

System Documentation

What problem does it solve?

Tiered KV caching for vLLM workloads enables production-grade long-context inference by offloading KV data to CPU DRAM, NVMe, or disaggregated backends, reducing re-fetches and latency.

Core Features & Use Cases

  • Supports Native CPU offload, LMCache DRAM+NVMe, NixlConnector for disaggregated prefill, MooncakeConnector for RDMA, and MultiConnector to compose backends.
  • Provides sizing guidance and version-gated recommendations to optimize throughput and hit rates for long-context prompts.
  • Use case: operate a multi-GPU vLLM service with high-context sessions and daily re-use of prompts across requests.

Quick Start

Configure a vLLM deployment to enable tiered KV caching and verify a small benchmark.

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: vllm-caching
Download link: https://github.com/air-gapped/skills/archive/main.zip#vllm-caching

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.