clm_22bb96006d8447c482c687ea057774f2_r1

IX2-0462

① SA Source

• Source: 開啟完整 SA 文章
• Section: Disaggregated Prefill
• Line hint: 672

Context Before

Source: DistServe ↗

Disaggregation also enables independent scaling and optimization of each phase. With separate nodes, each phase can be tuned independently: different parallelism strategies, different batch sizes, and different memory allocation ratios. The ratio of prefill to decode nodes can also be matched to the workload’s input-output length ratio. For instance, prefill-dominated workloads (long input, short output e.g., summarization, RAG, agentic coding with large context windows) allocate more prefill instances. Decode-dominated workloads (short input, long output e.g., chain-of-thought reasoning, long-form generation) allocate more decode instances. Workloads with high cache hit rates also tend toward more decode, since reused KV cache entries from shared system prompts or multi-turn conversation history skip prefill entirely.

Evidence

For a model like DeepSeek R1 with 61 layers and FP8 KV cache, an 8192-token prefill produces roughly 500MB of KV data that must cross the network, adding directly to TTFT

Context After

Source: Github ↗

② Atomic Claim

以 61 layers、FP8 KV cache 的 DeepSeek R1 為例，8192-token prefill 會產生約 500MB KV data，需要跨網路傳輸，因此直接增加 TTFT。

• Epistemic Mode: ASSERTED
• Mapping Status: COMPLETE

③ Semantic Frame

{
  "attribute": "LATENCY",
  "context_nodes": [
    {
      "id": "04_knowledge_base/KV cache",
      "label": "KV cache"
    },
    {
      "id": "02_companies/DeepSeek",
      "label": "DeepSeek"
    },
    {
      "id": "04_knowledge_base/Prefill",
      "label": "prefill"
    },
    {
      "id": "04_knowledge_base/Time to First Token",
      "label": "TTFT"
    }
  ],
  "entity": {
    "id": "04_knowledge_base/FP8",
    "label": "FP8"
  },
  "frame_type": "ATTRIBUTE",
  "qualifiers": {
    "condition_text": null,
    "numeric_mentions": [
      "61 layers",
      "8192",
      "500MB"
    ],
    "temporal_mentions": []
  },
  "value": {
    "numeric_mentions": [
      "61 layers",
      "8192",
      "500MB"
    ],
    "value_text": "以 61 layers、FP8 KV cache 的 DeepSeek R1 為例，8192-token prefill 會產生約 500MB KV data，需要跨網路傳輸，因此直接增加 TTFT。"
  }
}

④ Canonical Entity Mapping

Role	Surface Label	Canonical Target
entity	FP8	FP8
context_0	KV cache	04_knowledge_base/KV cache
context_1	DeepSeek	DeepSeek
context_2	prefill	Prefill
context_3	TTFT	04_knowledge_base/Time to First Token

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。