SA260324R-0095_r1

SA260324R-0095

① SA Source

• Source: 開啟完整 SA 文章
• Section: “
• Line hint: 103

Context Before

Evidence

One way LPUs can improve decode phase latencies is by applying the Attention FFN Disaggregation (§9§) technique, introduced in MegaScale-Infer and Step-3.

Context After

② Atomic Claim

LPU 改善 decode 階段延遲的一種方式，是採用 Attention FFN Disaggregation（AFD）；此技術曾在 MegaScale-Infer 與 Step-3 中提出。

• Epistemic Mode: ASSERTED
• Mapping Status: PARTIAL

③ Semantic Frame

{
  "frame_type": "NARY_RELATION",
  "participants": [
    {
      "node": {
        "id": "04_knowledge_base/Groq LPU",
        "label": "LPU"
      },
      "role": "accelerator"
    },
    {
      "node": {
        "id": "04_knowledge_base/Decode",
        "label": "decode"
      },
      "role": "target_phase"
    },
    {
      "node": {
        "id": "04_knowledge_base/Attention-FFN Disaggregation (AFD)",
        "label": "AFD"
      },
      "role": "technique"
    },
    {
      "node": {
        "id": "04_knowledge_base/MegaScale-Infer",
        "label": "MegaScale-Infer"
      },
      "role": "introduced_in"
    }
  ],
  "qualifiers": {
    "condition_text": "Step-3 亦被原文列為提出此技術的來源，但目前沒有 canonical mapping",
    "numeric_mentions": [],
    "temporal_mentions": []
  },
  "relation_type": "APPLIES_TECHNIQUE_TO_IMPROVE_LATENCY"
}

④ Canonical Entity Mapping

Role	Surface Label	Canonical Target
accelerator	LPU	04_knowledge_base/Groq LPU
target_phase	decode	Decode
technique	AFD	04_knowledge_base/Attention-FFN Disaggregation (AFD)
introduced_in	MegaScale-Infer	MegaScale-Infer

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。