IX2-0222
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
By contrast, with narrower EP and more DP replicas (e.g., 5xDEP8), each of the 5 replicas needs its own full copy of the 670B weights, that’s 5×670B = 3.35T of redundant weight loading across the system
Context After
Source: SemiAnalysis InferenceX ↗
② Atomic Claim
相較之下，若使用較窄的 EP 與更多 DP replicas（例如 5×DEP8），5 個 replicas 每個都需要完整 670B weights，因此整個系統會重複載入 5×670B＝3.35T 的權重。
- Epistemic Mode:
HYPOTHETICAL - Mapping Status:
PARTIAL
③ Semantic Frame
{
"attribute": "REDUNDANT_WEIGHT_LOADING",
"context_nodes": [],
"entity": {
"id": "04_knowledge_base/Expert Parallelism",
"label": "EP"
},
"frame_type": "ATTRIBUTE",
"qualifiers": {
"condition_text": "narrower EP with more DP replicas (e.g. 5xDEP8)",
"numeric_mentions": [
"5",
"670B",
"3.35T"
],
"temporal_mentions": []
},
"value": {
"numeric_mentions": [
"5",
"670B",
"3.35T"
],
"value_text": "5×670B = 3.35T redundant weight loading"
}
}④ Canonical Entity Mapping
| Role | Surface Label | Canonical Target |
|---|---|---|
| entity | EP | 04_knowledge_base/Expert Parallelism |
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。