SA260324R-0130_r1

SA260324R-0130

① SA Source

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

Context Before

For LPUs, deploying a draft model or MTP layers is quite different from applying AFD. FFNs are stateless, while draft models and

Evidence

MTP layers require dynamic KV cache loading.

Context After

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

MTP layer 需要動態載入 KV cache。

• Epistemic Mode: ASSERTED
• Mapping Status: COMPLETE

③ Semantic Frame

{
  "additional_nodes": [],
  "frame_type": "RELATION",
  "object": {
    "id": "04_knowledge_base/KV cache",
    "label": "KV cache"
  },
  "predicate": "REQUIRES_DYNAMIC_LOADING_OF",
  "qualifiers": {
    "condition_text": null,
    "numeric_mentions": [],
    "temporal_mentions": []
  },
  "subject": {
    "id": "04_knowledge_base/Multi-Token Prediction",
    "label": "MTP"
  }
}

④ Canonical Entity Mapping

Role	Surface Label	Canonical Target
subject	MTP	04_knowledge_base/Multi-Token Prediction
object	KV cache	04_knowledge_base/KV cache

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。