clm_39e4092bb6a74b318c75ec8c17607259_r1

IX2-0220

① SA Source

• Source: 開啟完整 SA 文章
• Section: Nvidia Disagg Prefill and WideEP
• Line hint: 378

Context Before

SemiAnalysis InferenceX is free open source software and reader-supported. To receive new posts and support our work, consider becoming a free or paid subscriber.

Subscribed

Evidence

With wide EP (e.g., DEP32), 32 GPUs collectively hold and load the 670B weights once, each loading only its shard (~21B)

Context After

Source: SemiAnalysis InferenceX ↗

② Atomic Claim

使用 wide EP（例如 DEP32）時，32 顆 GPUs 共同持有並只載入一次 670B weights，每顆只需載入自己的 shard，約 21B。

• Epistemic Mode: ASSERTED
• Mapping Status: COMPLETE

③ Semantic Frame

{
  "attribute": "COUNT",
  "context_nodes": [
    {
      "id": "04_knowledge_base/GPU",
      "label": "GPUs"
    }
  ],
  "entity": {
    "id": "04_knowledge_base/Expert Parallelism",
    "label": "wide EP"
  },
  "frame_type": "ATTRIBUTE",
  "qualifiers": {
    "condition_text": null,
    "numeric_mentions": [
      "32",
      "670B",
      "21B"
    ],
    "temporal_mentions": []
  },
  "value": {
    "numeric_mentions": [
      "32",
      "670B",
      "21B"
    ],
    "value_text": "使用 wide EP（例如 DEP32）時，32 顆 GPUs 共同持有並只載入一次 670B weights，每顆只需載入自己的 shard，約 21B。"
  }
}

④ Canonical Entity Mapping

Role	Surface Label	Canonical Target
entity	wide EP	04_knowledge_base/Expert Parallelism
context_0	GPUs	GPU

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。