SA260324R-0114
① SA Source
- Source: 開啟完整 SA 文章
- Section:
GPU and LPU Integration: Attention FFN Disaggregation (AFD) - Line hint:
107
Context Before
Source: SemiAnalysis
Evidence
Comparing the two operations, we see attention is stateful due to dynamic KV cache loading patterns, whereas FFN is stateless since the computation only depends on the token inputs. Thus, we disaggregate the computation of attention and FFN. We map attention computations to GPUs, which handle dynamic workloads well. For FFNs, we map them to LPUs, since LPU architecture is inherently deterministic and benefits from static compute workloads.
Context After
Source: SemiAnalysis, MegaScale-Infer
② Atomic Claim
attention computation 會被分配到 GPU，因為 GPU 擅長處理 dynamic workload。
- Epistemic Mode:
ASSERTED - Mapping Status:
PARTIAL
③ Semantic Frame
{
"attribute": "WORKLOAD_SUITABILITY",
"context_nodes": [],
"entity": {
"id": "04_knowledge_base/GPU",
"label": "GPU"
},
"frame_type": "ATTRIBUTE",
"qualifiers": {
"condition_text": "attention 在此 claim 沒有 canonical entity",
"numeric_mentions": [],
"temporal_mentions": []
},
"value": {
"numeric_mentions": [],
"value_text": "擅長處理 dynamic workload，因此承擔 attention computation。"
}
}④ Canonical Entity Mapping
| Role | Surface Label | Canonical Target |
|---|---|---|
| entity | GPU | GPU |
⑤ V2 Draft（需要修正時直接編輯）
Inline repair
只編輯以下三個 Draft 區塊。完成後將
review_action設為REQUEST_REPAIR，再手動執行 Repair Preview。
執行腳本前，Draft 不會寫入 Claim Store 或 Graph。
V2 Atomic Claim Draft
attention computation 會被分配到 GPU，因為 GPU 擅長處理 dynamic workload。
V2 Semantic Frame Draft
{
"attribute": "WORKLOAD_SUITABILITY",
"context_nodes": [],
"entity": {
"id": "04_knowledge_base/GPU",
"label": "GPU"
},
"frame_type": "ATTRIBUTE",
"qualifiers": {
"condition_text": "attention 在此 claim 沒有 canonical entity",
"numeric_mentions": [],
"temporal_mentions": []
},
"value": {
"numeric_mentions": [],
"value_text": "擅長處理 dynamic workload，因此承擔 attention computation。"
}
}V2 Revision Metadata Draft
{
"epistemic_mode": "ASSERTED",
"mapping_status": "PARTIAL"
}⑥ 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。