NIEK2-0097
① SA Source
- Source: 開啟完整 SA 文章
- Section:
GPU and LPU Integration: Attention FFN Disaggregation (AFD) - Line hint:
87
Context Before
Source: Nvidia
Now with an understanding of what LPUs are good for we can understand how they fit into inference setups. NVIDIA introduced LPUs to improve the performance of high interactivity scenarios. In those scenarios, LPUs can leverage their low-latency capabilities to improve the decode phase latencies. One way LPUs can improve decode phase latencies is by applying the Attention FFN Disaggregation (AFD) technique, introduced in MegaScale-Infer ↗ and Step-3 ↗.
Evidence
As we explained in our InferenceX article ↗, LLM inference involves two phases: prefill and decode. Prefill processes the full input context: It is compute-intensive, which is suitable for GPUs. On the other hand, decode predicts new tokens and is memory-bounded. Decode is latency-sensitive because the model predicts new tokens one by one, and LPU’s high SRAM bandwidth and low-latency capabilities can help accelerate this iterative process.
Context After
Source: SemiAnalysis
② Atomic Claim
如 InferenceX 文章所述，LLM 推論分為 prefill 與 decode 兩個階段。
- Epistemic Mode:
ASSERTED - Mapping Status:
PARTIAL
③ Semantic Frame
{
"frame_type": "NARY_RELATION",
"participants": [
{
"node": {
"id": "04_knowledge_base/InferenceX",
"label": "InferenceX"
},
"role": "subject"
},
{
"node": {
"id": "04_knowledge_base/Large language model",
"label": "LLM"
},
"role": "participant"
},
{
"node": {
"id": "04_knowledge_base/Prefill",
"label": "prefill"
},
"role": "participant"
},
{
"node": {
"id": "04_knowledge_base/Decode",
"label": "decode"
},
"role": "participant"
}
],
"qualifiers": {
"condition_text": null,
"numeric_mentions": [],
"temporal_mentions": []
},
"relation_type": "PARTITIONS_INTO"
}④ Canonical Entity Mapping
| Role | Surface Label | Canonical Target |
|---|---|---|
| subject | InferenceX | InferenceX |
| participant | LLM | 04_knowledge_base/Large language model |
| participant | prefill | Prefill |
| participant | decode | Decode |
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。