quiz.json

{
  "lesson": "18-vllm-production-stack-lmcache",
  "title": "vLLM Production Stack with LMCache KV Offloading",
  "questions": [
    {
      "stage": "pre",
      "question": "What problem does LMCache primarily address in a vLLM deployment?",
      "options": [
        "KV cache pressure in HBM causing preemption and re-prefill of the same prefixes",
        "Tokenizer GIL contention",
        "Network egress filtering",
        "Cold-start image pull"
      ],
      "correct": 0,
      "explanation": ""
    },
    {
      "stage": "check",
      "question": "What vLLM API introduced pluggable KV cache backends?",
      "options": [
        "Prefix-caching flag",
        "PagedAttention v2",
        "Connector API in vLLM v0.9.0",
        "ChunkedPrefill API"
      ],
      "correct": 2,
      "explanation": ""
    },
    {
      "stage": "check",
      "question": "What does the vLLM 0.11.0 (January 2026) release add to the KV offload path?",
      "options": [
        "An asynchronous offload path so the engine does not block on offload in the common case",
        "Synchronous-only offload",
        "Mandatory FP8 KV cache",
        "Removal of LMCache support"
      ],
      "correct": 0,
      "explanation": ""
    },
    {
      "stage": "check",
      "question": "When should you pick LMCache over native CPU offload?",
      "options": [
        "When multiple engines share prefixes across tenants, LoRA variants, or repeated RAG context, so cross-engine reuse pays",
        "When you want to disable KV caching entirely",
        "When a single engine has HBM pressure and no prefix sharing",
        "When you are running on CPU only"
      ],
      "correct": 0,
      "explanation": ""
    },
    {
      "stage": "post",
      "question": "What happens to LMCache benefit when KV footprint stays well below HBM?",
      "options": [
        "LMCache automatically disables",
        "It still doubles throughput",
        "Configs match baseline with roughly 3-5% overhead and no real benefit",
        "Engine crashes"
      ],
      "correct": 2,
      "explanation": ""
    },
    {
      "stage": "post",
      "question": "Why does LMCache compose with disaggregated serving (Phase 17 · 17)?",
      "options": [
        "Because LMCache replaces NIXL",
        "KV transferred from prefill to decode lands in LMCache; later queries can pull from LMCache and skip prefill, so the cache-aware router can pick an engine whose local or LMCache-shared cache matches",
        "Because LMCache runs on the same GPU as the engine",
        "It does not — they are mutually exclusive"
      ],
      "correct": 1,
      "explanation": ""
    }
  ]
}