模块框图

flowchart TB
    subgraph Assets[模型资产层]
        GGUF[GGUF blob]
        TOK[Tokenizer 导出]
        TENSOR[Tensor payload 导出]
        IMAGE[Q25TOKRT image]
        GGUF --> TOK --> IMAGE
        GGUF --> TENSOR
    end

    subgraph Host[Host 计算层]
        REF[量化解码与参考 forward]
        GOLDEN[golden / fixture manifest]
        TENSOR --> REF --> GOLDEN
    end

    subgraph PL[HLS / PL 算子层]
        EMB[Embedding / RMSNorm]
        QKV[QKV + RoPE]
        ATTN[Attention core]
        MLP[MLP / matvec]
        LM[LM head + top-k]
        GOLDEN --> EMB
        GOLDEN --> QKV
        GOLDEN --> ATTN
        GOLDEN --> MLP
        GOLDEN --> LM
    end

    subgraph PS[PS / A53 层]
        UART[UART command parser]
        BPE[Tokenizer C API]
        STUB[P4 runtime command stub]
        IMAGE --> BPE
        UART --> BPE
        UART --> STUB
    end

    subgraph Board[ZCU106 执行层]
        BUILD[XSA / BSP / ELF / SD]
        CAPTURE[UART capture]
        CHECK[Validator / archive]
        UART --> BUILD --> CAPTURE --> CHECK
    end