llama.cpp/src/models/ernie4-5-moe.cpp

#include "models.h"

std::unique_ptr<llm_graph_context> llama_model_ernie4_5_moe::build_arch_graph(const llm_graph_params & params) const {
    return std::make_unique<graph>(*this, params);
}

llama_model_ernie4_5_moe::graph::graph(const llama_model & model, const llm_graph_params & params) :
    llm_graph_context(params) {
    const int64_t n_embd_head = hparams.n_embd_head_v();

    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k());
    GGML_ASSERT(n_embd_head == n_rot);

    ggml_tensor * cur;
    ggml_tensor * inpL;

    inpL = build_inp_embd(model.tok_embd);

    // inp_pos - contains the positions
    ggml_tensor * inp_pos = build_inp_pos();

    auto * inp_attn = build_attn_inp_kv();

    ggml_tensor * inp_out_ids = build_inp_out_ids();

    GGML_ASSERT(hparams.n_moe_layer_step > 0 && "Ernie 4.5 MoE requires n_moe_layer_step > 0");
    for (int il = 0; il < n_layer; ++il) {
        ggml_tensor * inpSA = inpL;
        // norm
        {
            cur = build_norm(inpL, model.layers[il].attn_norm, NULL, LLM_NORM_RMS, il);
            cb(cur, "attn_norm", il);
        }
        // self-attention
        {
            // compute Q and K and RoPE them
            auto [Qcur, Kcur, Vcur] = build_qkv(model.layers[il], cur,
                    n_embd_head, n_head, n_head_kv, il);

            Qcur = ggml_rope_ext(ctx0, Qcur, inp_pos, nullptr, n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
                                 ext_factor, attn_factor, beta_fast, beta_slow);

            Kcur = ggml_rope_ext(ctx0, Kcur, inp_pos, nullptr, n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
                                 ext_factor, attn_factor, beta_fast, beta_slow);

            cb(Qcur, "Qcur", il);
            cb(Kcur, "Kcur", il);
            cb(Vcur, "Vcur", il);

            cur = build_attn(inp_attn,
                    model.layers[il].wo, NULL, model.layers[il].wo_s,
                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f / sqrtf(float(n_embd_head)), il);
            cb(cur, "attn_out", il);
        }
        if (il == n_layer - 1 && inp_out_ids) {
            cur   = ggml_get_rows(ctx0, cur, inp_out_ids);
            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);
        }
        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);
        cb(ffn_inp, "ffn_inp", il);

        // feed-forward network
        bool is_moe_layer =
            static_cast<uint32_t>(il) >= hparams.n_layer_dense_lead && (il + 1) % hparams.n_moe_layer_step == 0;

        if (!is_moe_layer) {
            cur = build_norm(ffn_inp, model.layers[il].ffn_norm, NULL, LLM_NORM_RMS, il);
            cb(cur, "ffn_norm", il);

            cur = build_ffn(cur,
                    model.layers[il].ffn_up, NULL, NULL,
                    model.layers[il].ffn_gate, NULL, NULL,
                    model.layers[il].ffn_down, NULL, NULL,
                    NULL, LLM_FFN_SILU, LLM_FFN_PAR, il);
            cb(cur, "ffn_out", il);
        } else {
            // MoE branch
            cur = build_norm(ffn_inp, model.layers[il].ffn_norm, NULL, LLM_NORM_RMS, il);
            cb(cur, "ffn_norm", il);

            ggml_tensor * moe_out = build_moe_ffn(cur,
                                        model.layers[il].ffn_gate_inp,
                                        model.layers[il].ffn_up_exps,
                                        model.layers[il].ffn_gate_exps,
                                        model.layers[il].ffn_down_exps,
                                        model.layers[il].ffn_exp_probs_b,
                                        n_expert, n_expert_used,
                                        LLM_FFN_SILU, true,
                                        hparams.expert_weights_scale,
                                        LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
                                        il);
            cb(moe_out, "ffn_moe_out", il);

            // Shared expert (if present)
            if (hparams.n_ff_shexp > 0) {
                ggml_tensor * ffn_shexp =
                    build_ffn(cur,
                        model.layers[il].ffn_up_shexp, NULL, NULL,
                        model.layers[il].ffn_gate_shexp, NULL, NULL,
                        model.layers[il].ffn_down_shexp, NULL, NULL,
                        NULL, LLM_FFN_SILU, LLM_FFN_PAR, il);
                cb(ffn_shexp, "ffn_shexp", il);

                cur = ggml_add(ctx0, moe_out, ffn_shexp);
            } else {
                cur = moe_out;
            }
            cb(cur, "ffn_out", il);
        }
        cur = ggml_add(ctx0, cur, ffn_inp);
        cb(cur, "ffn_out", il);

        cur = build_cvec(cur, il);
        cb(cur, "l_out", il);

        // input for next layer
        inpL = cur;
    }
    cur = inpL;

    cur = build_norm(cur, model.output_norm, NULL, LLM_NORM_RMS, -1);

    cb(cur, "result_norm", -1);
    res->t_embd = cur;

    // lm_head
    cur = build_lora_mm(model.output, cur, model.output_s);

    cb(cur, "result_output", -1);
    res->t_logits = cur;

    ggml_build_forward_expand(gf, cur);
}