SA260324R-0133_r1

SA260324R-0133

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 MTP layers require dynamic KV cache loading. Each FFN is around hundreds of megabytes, whereas draft models and MTP layers take up tens of gigabytes.

Evidence

To support this memory usage, LPUs can access up to 256 GB of DDR5 per Fabric Expansion Logic FPGAs on the LPX compute tray.

Context After

② Atomic Claim

為了支援這種 memory usage，LPU 可透過 LPX compute tray 上每顆 Fabric Expansion Logic FPGA 存取最高 256GB 的 DDR5。

• Epistemic Mode: ASSERTED
• Mapping Status: COMPLETE

③ Semantic Frame

{
  "frame_type": "NARY_RELATION",
  "participants": [
    {
      "node": {
        "id": "04_knowledge_base/Groq LPU",
        "label": "LPU"
      },
      "role": "compute_device"
    },
    {
      "node": {
        "id": "04_knowledge_base/DDR5",
        "label": "DDR5"
      },
      "role": "memory"
    },
    {
      "node": {
        "id": "04_knowledge_base/FPGA",
        "label": "FPGA"
      },
      "role": "fabric_expansion_logic"
    },
    {
      "node": {
        "id": "04_knowledge_base/Groq LPX",
        "label": "LPX"
      },
      "role": "rack_system"
    },
    {
      "node": {
        "id": "04_knowledge_base/Compute Tray",
        "label": "compute tray"
      },
      "role": "tray"
    }
  ],
  "qualifiers": {
    "condition_text": "每顆 Fabric Expansion Logic FPGA",
    "numeric_mentions": [
      "256GB"
    ],
    "temporal_mentions": []
  },
  "relation_type": "ACCESSES_DDR5_THROUGH_FPGA_ON_COMPUTE_TRAY"
}

④ Canonical Entity Mapping

Role	Surface Label	Canonical Target
compute_device	LPU	04_knowledge_base/Groq LPU
memory	DDR5	DDR5
fabric_expansion_logic	FPGA	FPGA
rack_system	LPX	04_knowledge_base/Groq LPX
tray	compute tray	04_knowledge_base/Compute Tray

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。