clm_f3a5942381fd413196353aac8ea89c13_r1

IX2-0396

① SA Source

• Source: 開啟完整 SA 文章
• Section: WideEP
• Line hint: 634

Context Before

By now, most frontier AI labs employ Mixture of Experts (MoE) model architectures as opposed to dense. In MoE architectures, only a subset of “experts” are activated for each token. For instance, DeepSeek R1 has 671B total parameters, but only 37B active parameters. Specifically, DeepSeek R1 has 256 routed experts (and 1 shared expert) with each token being routed to 8 distinct experts. This architecture lends itself naturally to expert parallelism (EP), which evenly distributes expert weights across some number of GPUs.

Consider serving DeepSeek R1 on a single 8-GPU server. At 671B parameters, some form of parallelism is required to fit the model across available HBM. The naive approach is tensor parallelism (TP), which shards every weight matrix across all GPUs. This works well for dense models but ignores the sparse activation pattern of MoE. With TP=8, each expert’s weights are sharded across all 8 GPUs, meaning every expert activation requires an all-reduce across all GPUs & the reduction dims of the GEMM is smaller leading to lower arithmetic intensity, even though only 8 of 256 experts activate per token. TP treats each expert like a dense layer, paying full cross-GPU communication cost while the model’s sparsity goes unexploited.

Evidence

Each GPU holds approximately 1/8th of the expert weights plus a full replica of the non-expert weights (attention projections, embeddings, normalization, and the shared expert)

Context After

The forward pass proceeds in two phases per layer. During attention, each GPU acts as an independent data-parallel rank, processing its own subset of requests using its replicated non-expert weights, no inter-GPU communication is needed. During the MoE phase, a lightweight router determines which experts each token requires, and tokens are dispatched to the appropriate GPUs via all-to-all communication. Each GPU executes its local experts on only the tokens routed to it, and results are returned via a second all-to-all.

② Atomic Claim

每顆 GPU 約持有 1/8 的 expert weights，外加完整複製一份 non-expert weights，例如 attention projections、embeddings、normalization 與 shared expert。

• Epistemic Mode: ASSERTED
• Mapping Status: COMPLETE

③ Semantic Frame

{
  "attribute": "COUNT",
  "context_nodes": [],
  "entity": {
    "id": "04_knowledge_base/GPU",
    "label": "GPU"
  },
  "frame_type": "ATTRIBUTE",
  "qualifiers": {
    "condition_text": null,
    "numeric_mentions": [
      "1",
      "8"
    ],
    "temporal_mentions": []
  },
  "value": {
    "numeric_mentions": [
      "1",
      "8"
    ],
    "value_text": "每顆 GPU 約持有 1/8 的 expert weights，外加完整複製一份 non-expert weights，例如 attention projections、embeddings、normalization 與 shared expert。"
  }
}

④ Canonical Entity Mapping

Role	Surface Label	Canonical Target
entity	GPU	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。