vshmidt/llama.cpp: LLM inference in C/C++

src/models/llama4.cpp
171 строка6 KB
Xuan-Son Nguyen
model: using single llm_build per arch (#21970)
16 апр 2026, 22:10
16 апр 2026, 22:104fbdabd
#include "models.h"

template <bool iswa>
llm_build_llama4<iswa>::llm_build_llama4(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();

    // temperature tuning
    ggml_tensor * inp_attn_scale = nullptr;
    inp_attn_scale = build_inp_attn_scale();

    using inp_attn_type = std::conditional_t<iswa, llm_graph_input_attn_kv_iswa, llm_graph_input_attn_kv>;
    inp_attn_type * inp_attn = nullptr;

    if constexpr (iswa) {
        inp_attn = build_attn_inp_kv_iswa();
    } else {
        inp_attn = build_attn_inp_kv();
    }

    const float kq_scale = hparams.f_attention_scale == 0.0f ? 1.0f/sqrtf(float(n_embd_head)) : hparams.f_attention_scale;

    ggml_tensor * inp_out_ids = build_inp_out_ids();

    for (int il = 0; il < n_layer; ++il) {
        const float freq_base_l  = model.get_rope_freq_base (cparams, il);
        const float freq_scale_l = model.get_rope_freq_scale(cparams, il);

        ggml_tensor * inpSA = inpL;

        // This overlaps with SWA layers in current models, so get_rope_freq_base/scale may be superfluous
        const bool use_rope = hparams.n_no_rope_layer_step > 0 &&
                              (il + 1) % hparams.n_no_rope_layer_step != 0;

        // norm
        cur = build_norm(inpL,
                model.layers[il].attn_norm, NULL,
                LLM_NORM_RMS, il);
        cb(cur, "attn_norm", il);

        // self-attention
        {
            // rope freq factors for llama3; may return nullptr for llama2 and other models
            ggml_tensor * rope_factors = model.get_rope_factors(cparams, il);

            // 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);

            if (use_rope) {
                Qcur = ggml_rope_ext(
                        ctx0, Qcur, inp_pos, rope_factors,
                        n_rot, rope_type, n_ctx_orig, freq_base_l, freq_scale_l,
                        ext_factor, attn_factor, beta_fast, beta_slow
                        );

                Kcur = ggml_rope_ext(
                        ctx0, Kcur, inp_pos, rope_factors,
                        n_rot, rope_type, n_ctx_orig, freq_base_l, freq_scale_l,
                        ext_factor, attn_factor, beta_fast, beta_slow
                        );
            } else if (inp_attn_scale) {
                Qcur = ggml_mul(ctx0, Qcur, inp_attn_scale);
            }
            cb(Qcur, "Qcur", il);
            cb(Kcur, "Kcur", il);
            cb(Vcur, "Vcur", il);

            if (use_rope && hparams.use_kq_norm) {
                // Llama4TextL2Norm
                Qcur = ggml_rms_norm(ctx0, Qcur, hparams.f_norm_rms_eps);
                Kcur = ggml_rms_norm(ctx0, Kcur, hparams.f_norm_rms_eps);
                cb(Qcur, "Qcur_normed", il);
                cb(Kcur, "Kcur_normed", il);
            }
            cur = build_attn(inp_attn,
                    model.layers[il].wo, model.layers[il].bo, model.layers[il].wo_s,
                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, kq_scale, 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 (non-MoE)
        if (model.layers[il].ffn_gate_inp == nullptr) {
            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,   model.layers[il].ffn_up_b,   NULL,
                    model.layers[il].ffn_gate, model.layers[il].ffn_gate_b, NULL,
                    model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,
                    NULL,
                    LLM_FFN_SILU, LLM_FFN_PAR, il);
            cb(cur, "ffn_out", il);
        } else {
            ggml_tensor * ffn_inp_normed = 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(ffn_inp_normed,
                    model.layers[il].ffn_gate_inp,
                    model.layers[il].ffn_up_exps,
                    model.layers[il].ffn_gate_exps,
                    model.layers[il].ffn_down_exps,
                    nullptr,
                    n_expert, n_expert_used,
                    LLM_FFN_SILU, false,
                    hparams.expert_weights_scale,
                    LLAMA_EXPERT_GATING_FUNC_TYPE_SIGMOID,
                    il);

            // Shared experts
            ggml_tensor * shexp_out = build_ffn(ffn_inp_normed,
                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(shexp_out, "ffn_moe_shexp", il);

            cur = ggml_add(ctx0, moe_out, shexp_out);
            cb(cur, "ffn_moe_out_merged", 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);

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

    ggml_build_forward_expand(gf, cur);
}

// Explicit template instantiations
template struct llm_build_llama4<false>;
template struct llm_build_llama4<true>;