quiz.json

{
  "lesson": "05-eagle3-speculative-decoding",
  "title": "EAGLE-3 Speculative Decoding in Production",
  "questions": [
    {
      "stage": "pre",
      "question": "Why does speculative decoding exploit a gap that exists in plain decode?",
      "options": [
        "Decode is compute-bound, so adding more compute is free",
        "Decode does not benefit from batching",
        "Decode requires more network bandwidth than prefill",
        "Decode is memory-bound, so the GPU is mostly idle waiting on HBM reads of weights"
      ],
      "correct": 3,
      "explanation": ""
    },
    {
      "stage": "check",
      "question": "What does the acceptance rate alpha measure?",
      "options": [
        "Fraction of GPU memory used during decode",
        "Fraction of draft-proposed tokens accepted by the target model",
        "Latency overhead of the draft model",
        "Cache hit rate of the KV cache"
      ],
      "correct": 1,
      "explanation": ""
    },
    {
      "stage": "check",
      "question": "What changes in EAGLE-3 compared to EAGLE-2 that pushes alpha to roughly 0.6-0.8 on general chat?",
      "options": [
        "It uses a full-sized draft model of the same family",
        "It runs on CPU instead of GPU",
        "It removes the verify step entirely",
        "The draft head is trained on multiple target layers rather than just the last layer"
      ],
      "correct": 3,
      "explanation": ""
    },
    {
      "stage": "check",
      "question": "Below roughly what alpha does the lesson say speculative decoding becomes net negative at high concurrency on most 2026 hardware?",
      "options": [
        "0.95",
        "0.05",
        "0.55",
        "0.85"
      ],
      "correct": 2,
      "explanation": ""
    },
    {
      "stage": "post",
      "question": "Which metric should you watch most closely after flipping EAGLE-3 on, even if mean ITL drops?",
      "options": [
        "Cold-start time",
        "GPU memory utilization",
        "P99 ITL, because rejected-draft two-passes can serialize under full batch",
        "Mean E2E latency"
      ],
      "correct": 2,
      "explanation": ""
    },
    {
      "stage": "post",
      "question": "Why is speculative decoding opt-in (not default) in vLLM 2026 per the lesson?",
      "options": [
        "Acceptance rate depends on workload, and turning it on without measuring alpha is a production anti-pattern",
        "It only works on Blackwell GPUs",
        "It is incompatible with PagedAttention",
        "It requires a separate license"
      ],
      "correct": 0,
      "explanation": ""
    }
  ]
}