SA260324R-0128_r1

SA260324R-0128

① SA Source

• Source: 開啟完整 SA 文章
• Section: Speculative Decoding
• Line hint: 167

Context Before

Speculative decoding usually boosts output token per decode step by 1.5 to 2 tokens, depending on the draft model / MTP accuracy. With its low latency capabilities, LPUs can further increase the latency savings and improve throughput.

Evidence

For LPUs, deploying a draft model or MTP layers is quite different from applying AFD.

Context After

FFNs are stateless, while draft models and MTP layers require dynamic KV cache loading. Each FFN is around hundreds of megabytes, whereas draft models and MTP layers take up tens of gigabytes. To support this memory usage, LPUs can access up to 256 GB of DDR5 per Fabric Expansion Logic FPGAs on the LPX compute tray.

② Atomic Claim

對 LPU 而言，部署 draft model 或 MTP layer，和套用 AFD 有很大差異。

• Epistemic Mode: ASSERTED
• Mapping Status: PARTIAL

③ Semantic Frame

{
  "comparison_expression": "在 LPU 上部署 draft model 或 MTP layer，和套用 AFD 有很大差異。",
  "entities": [
    {
      "id": "04_knowledge_base/Groq LPU",
      "label": "LPU"
    },
    {
      "id": "04_knowledge_base/Multi-Token Prediction",
      "label": "MTP"
    },
    {
      "id": "04_knowledge_base/Attention-FFN Disaggregation (AFD)",
      "label": "AFD"
    }
  ],
  "frame_type": "COMPARISON",
  "metric": "DEPLOYMENT_CHARACTERISTICS",
  "operator": "DIFFERENT_FROM",
  "qualifiers": {
    "condition_text": "draft model 沒有 canonical entity",
    "numeric_mentions": [],
    "temporal_mentions": []
  }
}

④ Canonical Entity Mapping

Role	Surface Label	Canonical Target
entity	LPU	04_knowledge_base/Groq LPU
entity	MTP	04_knowledge_base/Multi-Token Prediction
entity	AFD	04_knowledge_base/Attention-FFN Disaggregation (AFD)

⑤ Human Review

請在 Properties 逐項確認：

• 原文 → Atomic Claim 是否忠實
• Atomic Claim → Semantic Frame 是否忠實
• Canonical Entity mapping 是否正確
• Epistemic mode 是否保留原文語氣
• 最後選擇 review_action

Review state

Markdown 內文不是正式 approval。只有 Apply bridge 寫入的 Decision Ledger event 才是正式決策。