SA260324R-0116_r1

SA260324R-0116

① SA Source

• Source: 開啟完整 SA 文章
• Section: GPU and LPU Integration: Attention FFN Disaggregation (AFD)
• Line hint: 131

Context Before

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.

Evidence

With AFD, token routing from GPUs to LPUs can become the bottleneck, especially under strict latency constraints.

Context After

The token routing flow involves two operations: dispatch and combine. In the dispatch step, we route each token to their top k experts with an All-to-All collective operation. After experts complete their computation, we perform the combine step, where the outputs are sent back to the source location with a reverse All-to-All collective, continuing the next layer’s computation.

② Atomic Claim

在 AFD 下，從 GPU 到 LPU 的 token routing 可能成為 bottleneck，尤其是在嚴格 latency constraint 下。

• Epistemic Mode: ASSERTED
• Mapping Status: COMPLETE

③ Semantic Frame

{
  "frame_type": "NARY_RELATION",
  "participants": [
    {
      "node": {
        "id": "04_knowledge_base/Attention-FFN Disaggregation (AFD)",
        "label": "AFD"
      },
      "role": "technique"
    },
    {
      "node": {
        "id": "04_knowledge_base/GPU",
        "label": "GPU"
      },
      "role": "source_device"
    },
    {
      "node": {
        "id": "04_knowledge_base/Groq LPU",
        "label": "LPU"
      },
      "role": "destination_device"
    }
  ],
  "qualifiers": {
    "condition_text": "尤其在嚴格 latency constraint 下",
    "numeric_mentions": [],
    "temporal_mentions": []
  },
  "relation_type": "TOKEN_ROUTING_CAN_BECOME_BOTTLENECK"
}

④ Canonical Entity Mapping

Role	Surface Label	Canonical Target
technique	AFD	04_knowledge_base/Attention-FFN Disaggregation (AFD)
source_device	GPU	GPU
destination_device	LPU	04_knowledge_base/Groq LPU

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。