SA260324R-0103_r1

SA260324R-0103

① SA Source

• Source: 開啟完整 SA 文章
• Section: GPU and LPU Integration: Attention FFN Disaggregation (AFD)
• Line hint: 111

Context Before

Attention and FFN are subsets of operations in a model. In a model forward pass, attention’s output feeds into a token router, and

Evidence

the token router assigns each token to k experts, where each expert is an FFN. Attention and FFN have very different performance properties.

Context After

During decode phase, the GPU utilization of attention barely improves when scaling batch size due to being bounded by loading KV cache. In contrast, the GPU utilization of FFN scales with batch size comparatively better.

② Atomic Claim

token router 會將每個 token 分配給 k 個 expert，而每個 expert 都是一個 FFN；Attention 與 FFN 的效能特性差異很大。

• Epistemic Mode: ASSERTED
• Mapping Status: PARTIAL

③ Semantic Frame

{
  "frame_type": "CLAIM_ONLY",
  "qualifiers": {
    "condition_text": null,
    "numeric_mentions": [
      "k"
    ],
    "temporal_mentions": []
  },
  "reason": "TOKEN_ROUTER_AND_ATTENTION_HAVE_NO_CANONICAL_ENTITY_IN_THIS_CLAIM",
  "statement": "token router 會將每個 token 分配給 k 個 expert，而每個 expert 都是一個 FFN；Attention 與 FFN 的效能特性差異很大。"
}

④ Canonical Entity Mapping

Role	Surface Label	Canonical Target
—	—	—

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。