clm_4a81c1cd8644466f8904a892a0ef7a11_r1

IX2-0329

① SA Source

• Source: 開啟完整 SA 文章
• Section: Multi Token Prediction (MTP)
• Line hint: 540

Context Before

Source: SemiAnalysis InferenceX ↗

At large batch sizes, the inference regime is less memory-bandwidth bound compared to for low batch sizes. Since speculative decoding (including MTP) works by trading excess compute for fewer memory-bound decoding steps, this extra verification work from speculative tokens may not fit cleanly into slack, resulting in smaller improvements at high batch sizes.

Evidence

In terms of cost, MTP can drive huge cost savings, in the below table, we see that DeepSeek-R1-0528 run on FP4 using Dynamo TRT costs $0.251 per million total tokens

Context After

Source: SemiAnalysis InferenceX ↗

② Atomic Claim

就成本而言，MTP 可帶來巨大節省；表中 DeepSeek-R1-0528 在 FP4 下使用 Dynamo TRT，每百萬 total tokens 成本約 0.251 美元。

• Epistemic Mode: ASSERTED
• Mapping Status: PARTIAL

③ Semantic Frame

{
  "additional_nodes": [
    {
      "id": "04_knowledge_base/Multi-Token Prediction",
      "label": "MTP"
    }
  ],
  "frame_type": "RELATION",
  "object": {
    "id": "04_knowledge_base/FP4",
    "label": "FP4"
  },
  "predicate": "USES",
  "qualifiers": {
    "condition_text": null,
    "numeric_mentions": [
      "0528",
      "0.251"
    ],
    "temporal_mentions": []
  },
  "subject": {
    "id": "02_companies/DeepSeek",
    "label": "DeepSeek"
  }
}

④ Canonical Entity Mapping

Role	Surface Label	Canonical Target
subject	DeepSeek	DeepSeek
object	FP4	FP4
additional_0	MTP	04_knowledge_base/Multi-Token Prediction

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。