llama-cpp-sys-4 0.3.1

#include "models.h"

void llama_model_step35::load_arch_hparams(llama_model_loader & ml) {
    ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, hparams.f_norm_rms_eps);

    hparams.swa_type = LLAMA_SWA_TYPE_STANDARD;

    // full_attention layer only use half of the RoPE dimensions
    hparams.n_rot_full = hparams.n_rot_full / 2;

    // MoE + SWA parameters
    ml.get_key(LLM_KV_EXPERT_FEED_FORWARD_LENGTH,        hparams.n_ff_exp);
    ml.get_key(LLM_KV_EXPERT_SHARED_FEED_FORWARD_LENGTH, hparams.n_ff_shexp, false);
    ml.get_key(LLM_KV_EXPERT_GATING_FUNC,                hparams.expert_gating_func, false);
    ml.get_key(LLM_KV_EXPERT_WEIGHTS_SCALE,              hparams.expert_weights_scale, false);
    ml.get_key(LLM_KV_EXPERT_WEIGHTS_NORM,               hparams.expert_weights_norm, false);

    // Step35 uses sigmoid gating by default (if not set in GGUF)
    if (hparams.expert_gating_func == LLAMA_EXPERT_GATING_FUNC_TYPE_NONE) {
        hparams.expert_gating_func = LLAMA_EXPERT_GATING_FUNC_TYPE_SIGMOID;
    }

    ml.get_key(LLM_KV_ATTENTION_SLIDING_WINDOW,  hparams.n_swa);
    ml.get_key(LLM_KV_ROPE_FREQ_BASE_SWA,        hparams.rope_freq_base_train_swa, false);

    ml.get_key_or_arr(LLM_KV_ATTENTION_SLIDING_WINDOW_PATTERN, hparams.is_swa_impl, hparams.n_layer);

    ml.get_key_or_arr(LLM_KV_SWIGLU_CLAMP_EXP,   hparams.swiglu_clamp_exp,   hparams.n_layer, false);
    ml.get_key_or_arr(LLM_KV_SWIGLU_CLAMP_SHEXP, hparams.swiglu_clamp_shexp, hparams.n_layer, false);

    // NextN/MTP (Step3p5): extra decoder block appended beyond the main stack.
    ml.get_key(LLM_KV_NEXTN_PREDICT_LAYERS, hparams.nextn_predict_layers, false);
    GGML_ASSERT(hparams.nextn_predict_layers < hparams.n_layer && "nextn_predict_layers must be < n_layer");

    switch (hparams.n_layer - hparams.nextn_predict_layers) {
        case 45: type = LLM_TYPE_196B_A11B; break;
        default: type = LLM_TYPE_UNKNOWN;
    }
}

void llama_model_step35::load_arch_tensors(llama_model_loader & ml) {
    LLAMA_LOAD_LOCALS;

    const uint32_t n_main = n_layer - hparams.nextn_predict_layers;
    const bool mtp_only   = (hparams.nextn_predict_layers > 0) &&
                            (ml.get_weight("blk.0.attn_norm.weight") == nullptr);
    // Trunk-only: the GGUF declares MTP layers in metadata but the actual MTP
    // tensors live in a separate file (e.g. user split target/draft). Mark
    // MTP tensors NOT_REQUIRED so the trunk loads cleanly.
    const std::string mtp_probe = "blk." + std::to_string(n_main) + ".nextn.eh_proj.weight";
    const bool trunk_only = (hparams.nextn_predict_layers > 0) &&
                            (ml.get_weight(mtp_probe.c_str()) == nullptr);
    const int trunk_flags = mtp_only  ? TENSOR_NOT_REQUIRED : 0;
    const int mtp_flags   = trunk_only ? TENSOR_NOT_REQUIRED : 0;

    tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);

    // output
    output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
    output      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, trunk_flags);

    // STEP35 supports per-layer partial RoPE dims; rope factors are stored as a single shared tensor
    // ("rope_freqs.weight") and ggml uses only the first (n_rot_l/2) entries per layer.
    uint32_t n_rot_max = 0;
    for (int i = 0; i < n_layer; ++i) {
        n_rot_max = std::max(n_rot_max, hparams.n_rot(i));
    }
    if (n_rot_max == 0) {
        n_rot_max = n_rot;
    }

    auto load_block_trunk = [&](int i, int flags) {
        auto & layer = layers[i];

        const uint32_t n_head_l      = hparams.n_head(i);
        const uint32_t n_embd_k_gqa  = hparams.n_embd_k_gqa(i);
        const uint32_t n_embd_v_gqa  = hparams.n_embd_v_gqa(i);

        layer.attn_norm   = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, flags);
        layer.attn_q_norm = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "weight", i), {n_embd_head_k}, TENSOR_NOT_REQUIRED);
        layer.attn_k_norm = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "weight", i), {n_embd_head_k}, TENSOR_NOT_REQUIRED);

        // optional rope factors (llama3) / longrope tensors
        if (hparams.rope_scaling_type_train == LLAMA_ROPE_SCALING_TYPE_LONGROPE) {
            layer.rope_long  = create_tensor(tn(LLM_TENSOR_ROPE_FACTORS_LONG,  "weight", i), {n_rot_max/2}, TENSOR_NOT_REQUIRED | (i != 0 ? TENSOR_DUPLICATED : 0));
            layer.rope_short = create_tensor(tn(LLM_TENSOR_ROPE_FACTORS_SHORT, "weight", i), {n_rot_max/2}, TENSOR_NOT_REQUIRED | (i != 0 ? TENSOR_DUPLICATED : 0));
        } else {
            layer.rope_freqs = create_tensor(tn(LLM_TENSOR_ROPE_FREQS, "weight", i), {n_rot_max/2}, TENSOR_NOT_REQUIRED | (i != 0 ? TENSOR_DUPLICATED : 0));
        }

        create_tensor_qkv(layer, i, n_embd, n_embd_head_k * n_head_l, n_embd_k_gqa, n_embd_v_gqa, flags);
        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd_head_v * n_head_l, n_embd}, flags);

        // head-wise attention gate (Step35 self_attn.g_proj)
        layer.wqkv_gate = create_tensor(tn(LLM_TENSOR_ATTN_GATE, "weight", i), {n_embd, n_head_l}, TENSOR_NOT_REQUIRED);

        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, flags);

        // dense MLP (leading dense blocks)
        layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd,   n_ff}, TENSOR_NOT_REQUIRED);
        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd}, TENSOR_NOT_REQUIRED);
        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff}, TENSOR_NOT_REQUIRED);

        // MoE routed experts + selection bias (router_bias)
        const int64_t n_ff_exp = hparams.n_ff_exp;
        layer.ffn_gate_inp      = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP,  "weight", i), {n_embd, n_expert}, TENSOR_NOT_REQUIRED);
        layer.ffn_gate_exps     = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), {n_embd, n_ff_exp,   n_expert}, TENSOR_NOT_REQUIRED);
        layer.ffn_down_exps     = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), {n_ff_exp,   n_embd, n_expert}, TENSOR_NOT_REQUIRED);
        layer.ffn_up_exps       = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS,   "weight", i), {n_embd, n_ff_exp,   n_expert}, TENSOR_NOT_REQUIRED);
        layer.ffn_exp_probs_b   = create_tensor(tn(LLM_TENSOR_FFN_EXP_PROBS_B, "bias", i), {n_expert}, TENSOR_NOT_REQUIRED);

        // shared expert MLP
        layer.ffn_gate_shexp = create_tensor(tn(LLM_TENSOR_FFN_GATE_SHEXP, "weight", i), {n_embd, hparams.n_ff_shexp}, TENSOR_NOT_REQUIRED);
        layer.ffn_up_shexp   = create_tensor(tn(LLM_TENSOR_FFN_UP_SHEXP,   "weight", i), {n_embd, hparams.n_ff_shexp}, TENSOR_NOT_REQUIRED);
        layer.ffn_down_shexp = create_tensor(tn(LLM_TENSOR_FFN_DOWN_SHEXP, "weight", i), {hparams.n_ff_shexp, n_embd}, TENSOR_NOT_REQUIRED);
    };

    auto load_block_mtp = [&](int i, bool is_first_mtp) {
        auto & layer = layers[i];

        const uint32_t n_head_l      = hparams.n_head(i);
        const uint32_t n_embd_k_gqa  = hparams.n_embd_k_gqa(i);
        const uint32_t n_embd_v_gqa  = hparams.n_embd_v_gqa(i);

        // The MTP block is a full Step3p5 decoder layer (mtp_block) plus the
        // NextN-specific wiring (enorm/hnorm/eh_proj + optional shared head).
        // `mtp_flags` becomes NOT_REQUIRED when the GGUF is trunk-only.
        //
        // Only the FIRST MTP block (i == n_main) is required for the
        // single-block MTP runtime; trailing MTP blocks are always tolerated
        // as missing so pruned GGUFs (block 0 only) load cleanly. Override
        // mtp_flags to NOT_REQUIRED for those.
        const int eff_mtp_flags = is_first_mtp ? mtp_flags : (mtp_flags | TENSOR_NOT_REQUIRED);

        layer.attn_norm   = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, eff_mtp_flags);
        layer.attn_q_norm = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "weight", i), {n_embd_head_k}, TENSOR_NOT_REQUIRED);
        layer.attn_k_norm = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "weight", i), {n_embd_head_k}, TENSOR_NOT_REQUIRED);

        if (hparams.rope_scaling_type_train == LLAMA_ROPE_SCALING_TYPE_LONGROPE) {
            layer.rope_long  = create_tensor(tn(LLM_TENSOR_ROPE_FACTORS_LONG,  "weight", i), {n_rot_max/2}, TENSOR_NOT_REQUIRED | TENSOR_DUPLICATED);
            layer.rope_short = create_tensor(tn(LLM_TENSOR_ROPE_FACTORS_SHORT, "weight", i), {n_rot_max/2}, TENSOR_NOT_REQUIRED | TENSOR_DUPLICATED);
        } else {
            layer.rope_freqs = create_tensor(tn(LLM_TENSOR_ROPE_FREQS, "weight", i), {n_rot_max/2}, TENSOR_NOT_REQUIRED | TENSOR_DUPLICATED);
        }

        create_tensor_qkv(layer, i, n_embd, n_embd_head_k * n_head_l, n_embd_k_gqa, n_embd_v_gqa, eff_mtp_flags);
        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd_head_v * n_head_l, n_embd}, eff_mtp_flags);

        layer.wqkv_gate = create_tensor(tn(LLM_TENSOR_ATTN_GATE, "weight", i), {n_embd, n_head_l}, TENSOR_NOT_REQUIRED);

        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, eff_mtp_flags);

        // dense MLP (leading dense blocks) — present if the MTP block isn't MoE
        layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd,   n_ff}, TENSOR_NOT_REQUIRED);
        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), {  n_ff, n_embd}, TENSOR_NOT_REQUIRED);
        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", i), {n_embd,   n_ff}, TENSOR_NOT_REQUIRED);

        // MoE routed experts + selection bias (router_bias)
        const int64_t n_ff_exp = hparams.n_ff_exp;
        layer.ffn_gate_inp      = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP,  "weight", i), {n_embd, n_expert}, TENSOR_NOT_REQUIRED);
        layer.ffn_gate_exps     = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), {n_embd, n_ff_exp,   n_expert}, TENSOR_NOT_REQUIRED);
        layer.ffn_down_exps     = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), {n_ff_exp,   n_embd, n_expert}, TENSOR_NOT_REQUIRED);
        layer.ffn_up_exps       = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS,   "weight", i), {n_embd, n_ff_exp,   n_expert}, TENSOR_NOT_REQUIRED);
        layer.ffn_exp_probs_b   = create_tensor(tn(LLM_TENSOR_FFN_EXP_PROBS_B, "bias", i), {n_expert}, TENSOR_NOT_REQUIRED);

        layer.ffn_gate_shexp = create_tensor(tn(LLM_TENSOR_FFN_GATE_SHEXP, "weight", i), {n_embd, hparams.n_ff_shexp}, TENSOR_NOT_REQUIRED);
        layer.ffn_up_shexp   = create_tensor(tn(LLM_TENSOR_FFN_UP_SHEXP,   "weight", i), {n_embd, hparams.n_ff_shexp}, TENSOR_NOT_REQUIRED);
        layer.ffn_down_shexp = create_tensor(tn(LLM_TENSOR_FFN_DOWN_SHEXP, "weight", i), {hparams.n_ff_shexp, n_embd}, TENSOR_NOT_REQUIRED);

        // NextN-specific tensors that define the MTP block.
        layer.nextn.eh_proj          = create_tensor(tn(LLM_TENSOR_NEXTN_EH_PROJ,          "weight", i), { 2 * n_embd, n_embd }, eff_mtp_flags);
        layer.nextn.enorm            = create_tensor(tn(LLM_TENSOR_NEXTN_ENORM,            "weight", i), { n_embd },              eff_mtp_flags);
        layer.nextn.hnorm            = create_tensor(tn(LLM_TENSOR_NEXTN_HNORM,            "weight", i), { n_embd },              eff_mtp_flags);
        layer.nextn.embed_tokens     = create_tensor(tn(LLM_TENSOR_NEXTN_EMBED_TOKENS,     "weight", i), { n_embd, n_vocab },     TENSOR_NOT_REQUIRED);
        layer.nextn.shared_head_head = create_tensor(tn(LLM_TENSOR_NEXTN_SHARED_HEAD_HEAD, "weight", i), { n_embd, n_vocab },     TENSOR_NOT_REQUIRED);
        layer.nextn.shared_head_norm = create_tensor(tn(LLM_TENSOR_NEXTN_SHARED_HEAD_NORM, "weight", i), { n_embd },              TENSOR_NOT_REQUIRED);
    };

    for (int i = 0; i < (int) n_main; ++i) {
        load_block_trunk(i, trunk_flags);
    }
    // Only the first MTP block (i == n_main) is required at runtime — the
    // single-block-MTP graph in build_arch_graph always uses that one.
    // Trailing MTP blocks are loaded if present (so an un-pruned GGUF with
    // all MTP layers still works) but tolerated when absent via the pruning
    // path. See scripts/prune_step35_extra_mtp.py for the pruner.
    for (int i = (int) n_main; i < n_layer; ++i) {
        load_block_mtp(i, /*is_first_mtp=*/ i == (int) n_main);
    }
}

std::unique_ptr<llm_graph_context> llama_model_step35::build_arch_graph(const llm_graph_params & params) const {
    if (params.gtype == LLM_GRAPH_TYPE_DECODER_MTP) {
        return std::make_unique<graph_mtp>(*this, params);
    }
    return std::make_unique<graph>(*this, params);
}

llama_model_step35::graph::graph(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
    ggml_tensor * cur;
    ggml_tensor * inpL;

    inpL = build_inp_embd(model.tok_embd);
    ggml_tensor * inp_pos     = build_inp_pos();
    auto        * inp_attn    = build_attn_inp_kv_iswa();
    ggml_tensor * inp_out_ids = build_inp_out_ids();

    // MTP/NextN layers are loaded as extra decoder blocks but not executed in the main pass.
    const int n_transformer_layers = n_layer - (int) hparams.nextn_predict_layers;
    for (int il = 0; il < n_transformer_layers; ++il) {
        ggml_tensor * inpSA = inpL;

        const uint32_t n_head_l    = hparams.n_head(il);
        const uint32_t n_head_kv_l = hparams.n_head_kv(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);

        cur = inpL;

        // dump pre-attn RMSNorm input to pinpoint layer boundary issues
        cb(cur, "attn_norm_in", il);

        // self-attention
        {
            cur = build_norm(cur, model.layers[il].attn_norm, nullptr, LLM_NORM_RMS, il);
            cb(cur, "attn_norm", il);
            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);
            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);
            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);

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

            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head_k, n_head_l,    n_tokens);
            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head_k, n_head_kv_l, n_tokens);
            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head_v, n_head_kv_l, n_tokens);

            // Q/K per-head RMSNorm (Step35 q_norm / k_norm)
            if (model.layers[il].attn_q_norm) {
                Qcur = build_norm(Qcur, model.layers[il].attn_q_norm, nullptr, LLM_NORM_RMS, il);
                cb(Qcur, "Qcur_normed", il);
            }
            if (model.layers[il].attn_k_norm) {
                Kcur = build_norm(Kcur, model.layers[il].attn_k_norm, nullptr, LLM_NORM_RMS, il);
                cb(Kcur, "Kcur_normed", il);
            }

            // RoPE (partial rotary factors per layer)
            const bool is_swa = hparams.is_swa(il);
            ggml_tensor * rope_factors = is_swa ? nullptr : model.get_rope_factors(cparams, il);
            const int64_t n_rot_l = hparams.n_rot(il);
            Qcur = ggml_rope_ext(
                ctx0, Qcur, inp_pos, rope_factors,
                n_rot_l, 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_l, rope_type, n_ctx_orig, freq_base_l, freq_scale_l,
                ext_factor, attn_factor, beta_fast, beta_slow
            );
            cb(Qcur, "Qcur_pos", il);
            cb(Kcur, "Kcur_pos", il);

            const float kq_scale = 1.0f / sqrtf(float(n_embd_head_k));
            ggml_tensor * attn_out = build_attn(inp_attn,
                    nullptr, nullptr, nullptr,
                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, kq_scale, il);
            cb(attn_out, "attn_out", il);
            // head-wise attention gate: sigmoid(g_proj(x)) in torch
            if (model.layers[il].wqkv_gate) {
                ggml_tensor * gate = build_lora_mm(model.layers[il].wqkv_gate, cur); // [n_head_l, n_tokens]
                cb(gate, "attn_gate", il);

                gate = ggml_sigmoid(ctx0, gate);
                cb(gate, "attn_gate_sigmoid", il);

                // reshape + broadcast to [n_embd_head_v, n_head_l, n_tokens]
                ggml_tensor * attn_3d = ggml_reshape_3d(ctx0, attn_out, n_embd_head_v, n_head_l, n_tokens);
                ggml_tensor * gate_3d = ggml_reshape_3d(ctx0, gate,       1,          n_head_l, n_tokens);
                cb(gate_3d, "attn_gate_3d", il);

                attn_3d = ggml_mul(ctx0, attn_3d, gate_3d);
                cb(attn_3d, "attn_gated_3d", il);

                attn_out = ggml_reshape_2d(ctx0, attn_3d, n_embd_head_v * n_head_l, n_tokens);
                cb(attn_out, "attn_gated", il);
            }

            // output projection
            cur = build_lora_mm(model.layers[il].wo, attn_out, model.layers[il].wo_s);
            cb(cur, "attn_proj", il);
        }

        if (il == n_transformer_layers - 1 && inp_out_ids && cparams.embeddings_nextn_masked) {
            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);

        cur = build_norm(ffn_inp, model.layers[il].ffn_norm, nullptr, LLM_NORM_RMS, il);
        cb(cur, "ffn_norm", il);

        // feed-forward
        if (model.layers[il].ffn_gate_inp == nullptr) {
            // dense MLP
            cur = build_ffn(cur,
                    model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   nullptr,
                    model.layers[il].ffn_gate, model.layers[il].ffn_gate_b, nullptr,
                    model.layers[il].ffn_down, model.layers[il].ffn_down_b, nullptr,
                    nullptr,
                    LLM_FFN_SILU, LLM_FFN_PAR, il);
            cb(cur, "ffn_out", il);
        } else {
            // MoE routed experts
            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, hparams.expert_weights_norm,
                    hparams.expert_weights_scale,
                    (llama_expert_gating_func_type) hparams.expert_gating_func,
                    il);
            cb(moe_out, "ffn_moe_out", il);

            // shared expert MLP (always added on MoE layers in Step35)
            ggml_tensor * sh_out = build_ffn(cur,
                    model.layers[il].ffn_up_shexp,   nullptr, nullptr,
                    model.layers[il].ffn_gate_shexp, nullptr, nullptr,
                    model.layers[il].ffn_down_shexp, nullptr, nullptr,
                    nullptr,
                    LLM_FFN_SILU, LLM_FFN_PAR, il);
            cb(sh_out, "ffn_shared_out", il);

            cur = ggml_add(ctx0, moe_out, sh_out);
            cb(cur, "ffn_out", il);
        }
        cur = ggml_add(ctx0, cur, ffn_inp);

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

        // input for next layer
        inpL = cur;
    }

    cur = inpL;

    cb(cur, "h_nextn", -1);
    res->t_h_nextn = cur;

    if (!cparams.embeddings_nextn_masked && inp_out_ids) {
        cur = ggml_get_rows(ctx0, cur, inp_out_ids);
    }

    cur = build_norm(cur, model.output_norm, nullptr, LLM_NORM_RMS, -1);
    cb(cur, "result_norm", -1);
    res->t_embd = cur;

    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);
}

// LLM_GRAPH_TYPE_DECODER_MTP draft head for Step3p5 (MoE)
llama_model_step35::graph_mtp::graph_mtp(const llama_model & model, const llm_graph_params & params)
    : llm_graph_context(params) {
    GGML_ASSERT(hparams.nextn_predict_layers > 0 && "STEP35 MTP requires nextn_predict_layers > 0");

    // Single-block MTP only: always run the first trained MTP block (Qwen
    // MTP / vLLM single-MTP-layer style). Multi-block round-robin proved to
    // be a much deeper refactor than this PR justifies; the trailing MTP
    // blocks are loaded with TENSOR_NOT_REQUIRED so pruned GGUFs (with just
    // block 0) also work — see load_arch_tensors below and
    // scripts/prune_step35_extra_mtp.py.
    const int il       = (int) hparams.n_layer - (int) hparams.nextn_predict_layers;
    const auto & layer = model.layers[il];

    GGML_ASSERT(layer.nextn.eh_proj && "MTP block missing nextn.eh_proj");
    GGML_ASSERT(layer.nextn.enorm   && "MTP block missing nextn.enorm");
    GGML_ASSERT(layer.nextn.hnorm   && "MTP block missing nextn.hnorm");

    const uint32_t n_head_l    = hparams.n_head(il);
    const uint32_t n_head_kv_l = hparams.n_head_kv(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);

    auto inp = std::make_unique<llm_graph_input_embd>(hparams.n_embd);

    inp->tokens = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, n_tokens);
    ggml_set_input(inp->tokens);

    inp->embd = ggml_new_tensor_2d(ctx0, GGML_TYPE_F32, hparams.n_embd, n_tokens);
    ggml_set_input(inp->embd);
    ggml_set_name(inp->embd, "mtp_h_input");

    ggml_tensor * tok_embd_w = layer.nextn.embed_tokens ? layer.nextn.embed_tokens : model.tok_embd;

    ggml_tensor * h_input  = inp->embd;
    ggml_tensor * tok_embd = ggml_get_rows(ctx0, tok_embd_w, inp->tokens);
    cb(tok_embd, "mtp_tok_embd", il);

    res->add_input(std::move(inp));

    ggml_tensor * inp_pos  = build_inp_pos();
    auto        * inp_attn = build_attn_inp_kv_iswa();

    ggml_tensor * h_norm = build_norm(h_input, layer.nextn.hnorm, nullptr, LLM_NORM_RMS, il);
    cb(h_norm, "mtp_hnorm", il);

    ggml_tensor * e_norm = build_norm(tok_embd, layer.nextn.enorm, nullptr, LLM_NORM_RMS, il);
    cb(e_norm, "mtp_enorm", il);

    ggml_tensor * concat = ggml_concat(ctx0, e_norm, h_norm, /*dim=*/ 0);
    cb(concat, "mtp_concat", il);

    ggml_tensor * cur = build_lora_mm(layer.nextn.eh_proj, concat);
    cb(cur, "mtp_eh_proj", il);

    ggml_tensor * inpSA = cur;

    // mtp_block: full Step3p5 decoder layer (attention with optional head-wise gate, then MoE/dense FFN)
    cur = build_norm(cur, layer.attn_norm, nullptr, LLM_NORM_RMS, il);
    cb(cur, "mtp_attn_norm", il);

    ggml_tensor * Qcur = build_lora_mm(layer.wq, cur, layer.wq_s);
    ggml_tensor * Kcur = build_lora_mm(layer.wk, cur, layer.wk_s);
    ggml_tensor * Vcur = build_lora_mm(layer.wv, cur, layer.wv_s);
    cb(Qcur, "mtp_Qcur", il);
    cb(Kcur, "mtp_Kcur", il);
    cb(Vcur, "mtp_Vcur", il);

    Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head_k, n_head_l,    n_tokens);
    Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head_k, n_head_kv_l, n_tokens);
    Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head_v, n_head_kv_l, n_tokens);

    if (layer.attn_q_norm) {
        Qcur = build_norm(Qcur, layer.attn_q_norm, nullptr, LLM_NORM_RMS, il);
        cb(Qcur, "mtp_Qcur_normed", il);
    }
    if (layer.attn_k_norm) {
        Kcur = build_norm(Kcur, layer.attn_k_norm, nullptr, LLM_NORM_RMS, il);
        cb(Kcur, "mtp_Kcur_normed", il);
    }

    const bool    is_swa       = hparams.is_swa(il);
    ggml_tensor * rope_factors = is_swa ? nullptr : model.get_rope_factors(cparams, il);
    const int64_t n_rot_l      = hparams.n_rot(il);

    Qcur = ggml_rope_ext(
        ctx0, Qcur, inp_pos, rope_factors,
        n_rot_l, 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_l, rope_type, n_ctx_orig, freq_base_l, freq_scale_l,
        ext_factor, attn_factor, beta_fast, beta_slow);
    cb(Qcur, "mtp_Qcur_pos", il);
    cb(Kcur, "mtp_Kcur_pos", il);

    const float kq_scale = 1.0f / sqrtf(float(n_embd_head_k));
    ggml_tensor * attn_out = build_attn(inp_attn,
            nullptr, nullptr, nullptr,
            Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, kq_scale, il);
    cb(attn_out, "mtp_attn_out", il);

    // head-wise attention gate: sigmoid(g_proj(x))
    if (layer.wqkv_gate) {
        ggml_tensor * gate = build_lora_mm(layer.wqkv_gate, cur); // [n_head_l, n_tokens]
        cb(gate, "mtp_attn_gate", il);

        gate = ggml_sigmoid(ctx0, gate);
        cb(gate, "mtp_attn_gate_sigmoid", il);

        ggml_tensor * attn_3d = ggml_reshape_3d(ctx0, attn_out, n_embd_head_v, n_head_l, n_tokens);
        ggml_tensor * gate_3d = ggml_reshape_3d(ctx0, gate,       1,           n_head_l, n_tokens);
        cb(gate_3d, "mtp_attn_gate_3d", il);

        attn_3d = ggml_mul(ctx0, attn_3d, gate_3d);
        cb(attn_3d, "mtp_attn_gated_3d", il);

        attn_out = ggml_reshape_2d(ctx0, attn_3d, n_embd_head_v * n_head_l, n_tokens);
        cb(attn_out, "mtp_attn_gated", il);
    }

    cur = build_lora_mm(layer.wo, attn_out, layer.wo_s);
    cb(cur, "mtp_attn_proj", il);

    cur = ggml_add(ctx0, cur, inpSA);
    cb(cur, "mtp_attn_residual", il);

    ggml_tensor * ffn_inp = cur;
    cur = build_norm(cur, layer.ffn_norm, nullptr, LLM_NORM_RMS, il);
    cb(cur, "mtp_ffn_norm", il);

    // FFN: dense MLP or MoE (mirrors trunk path)
    if (layer.ffn_gate_inp == nullptr) {
        cur = build_ffn(cur,
                layer.ffn_up,   layer.ffn_up_b,   nullptr,
                layer.ffn_gate, layer.ffn_gate_b, nullptr,
                layer.ffn_down, layer.ffn_down_b, nullptr,
                nullptr,
                LLM_FFN_SILU, LLM_FFN_PAR, il);
        cb(cur, "mtp_ffn_out", il);
    } else {
        ggml_tensor * moe_out = build_moe_ffn(cur,
                layer.ffn_gate_inp,
                layer.ffn_up_exps,
                layer.ffn_gate_exps,
                layer.ffn_down_exps,
                layer.ffn_exp_probs_b,
                n_expert, n_expert_used,
                LLM_FFN_SILU, hparams.expert_weights_norm,
                hparams.expert_weights_scale,
                (llama_expert_gating_func_type) hparams.expert_gating_func,
                il);
        cb(moe_out, "mtp_ffn_moe_out", il);

        ggml_tensor * sh_out = build_ffn(cur,
                layer.ffn_up_shexp,   nullptr, nullptr,
                layer.ffn_gate_shexp, nullptr, nullptr,
                layer.ffn_down_shexp, nullptr, nullptr,
                nullptr,
                LLM_FFN_SILU, LLM_FFN_PAR, il);
        cb(sh_out, "mtp_ffn_shared_out", il);

        cur = ggml_add(ctx0, moe_out, sh_out);
        cb(cur, "mtp_ffn_out", il);
    }
    cur = ggml_add(ctx0, cur, ffn_inp);
    cb(cur, "mtp_post_ffn", il);

    // Pre-norm hidden state: used by the AR draft loop to seed the next MTP step.
    cb(cur, "h_nextn", -1);
    res->t_h_nextn = cur;

    ggml_tensor * head_norm_w = layer.nextn.shared_head_norm
            ? layer.nextn.shared_head_norm
            : model.output_norm;
    GGML_ASSERT(head_norm_w && "STEP35 MTP: missing both nextn.shared_head_norm and output_norm");
    cur = build_norm(cur, head_norm_w, nullptr, LLM_NORM_RMS, -1);
    cb(cur, "mtp_shared_head_norm", -1);

    ggml_tensor * head_w = layer.nextn.shared_head_head ? layer.nextn.shared_head_head : model.output;
    GGML_ASSERT(head_w && "STEP35 MTP: missing LM head (nextn.shared_head_head or model.output)");
    cur = build_lora_mm(head_w, cur);
    cb(cur, "result_output", -1);

    res->t_logits = cur;
    ggml_build_forward_expand(gf, cur);
}