#include "models.h"
#include "qwen35-specialized.h"
#include "llama-memory-recurrent.h"

#include <cstdio>
#include <cstdlib>
#include <cstring>

namespace {

// Resolve forced early-exit layer count for classifier embeddings.
// Priority: SCAMGUARDIAN_EARLY_EXIT_LAYERS_FILE (single int) then env LAYERS.
// Returns 0 when disabled (run full stack).
int scamguardian_early_exit_layers(int n_layer) {
    int early_layers = 0;
    const char * early_file = std::getenv("SCAMGUARDIAN_EARLY_EXIT_LAYERS_FILE");
    if (early_file != nullptr && early_file[0] != '\0') {
        FILE * f = std::fopen(early_file, "r");
        if (f) {
            int v = 0;
            if (std::fscanf(f, "%d", &v) == 1) {
                early_layers = v;
            }
            std::fclose(f);
        }
    } else {
        const char * early_layers_env = std::getenv("SCAMGUARDIAN_EARLY_EXIT_LAYERS");
        if (early_layers_env != nullptr && early_layers_env[0] != '\0') {
            early_layers = std::atoi(early_layers_env);
        }
    }
    if (early_layers > 0 && early_layers < n_layer) {
        return early_layers;
    }
    return 0;
}

int scamguardian_continuation_boundary(int n_layer) {
    const char * value = std::getenv("SCAMGUARDIAN_SUFFIX_BOUNDARY_LAYERS");
    if (value == nullptr || value[0] == '\0') {
        return 0;
    }
    const int boundary = std::atoi(value);
    return boundary > 0 && boundary < n_layer ? boundary : 0;
}

} // namespace

void llama_model_qwen35::load_arch_hparams(llama_model_loader & ml) {
    ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS,       hparams.f_norm_rms_eps);
    ml.get_key_or_arr(LLM_KV_ROPE_DIMENSION_SECTIONS,    hparams.rope_sections, 4, true);

    // Load linear attention (gated delta net) parameters
    ml.get_key(LLM_KV_SSM_CONV_KERNEL,    hparams.ssm_d_conv);
    ml.get_key(LLM_KV_SSM_INNER_SIZE,     hparams.ssm_d_inner);
    ml.get_key(LLM_KV_SSM_STATE_SIZE,     hparams.ssm_d_state);
    ml.get_key(LLM_KV_SSM_TIME_STEP_RANK, hparams.ssm_dt_rank);
    ml.get_key(LLM_KV_SSM_GROUP_COUNT,    hparams.ssm_n_group);

    // NextN/MTP (Qwen3.5/3.6): extra decoder block appended beyond the main stack
    ml.get_key(LLM_KV_NEXTN_PREDICT_LAYERS, hparams.n_layer_nextn, false);
    GGML_ASSERT(hparams.n_layer_nextn < hparams.n_layer_all && "n_layer_nextn must be < n_layer_impl");

    // Mark recurrent layers (linear attention layers). MTP layers are dense
    // attention-only and must be flagged non-recurrent.
    if (!ml.get_key_or_arr(LLM_KV_ATTENTION_RECURRENT_LAYERS, hparams.is_recr_impl, hparams.n_layer_all, false)) {
        uint32_t full_attn_interval = 4;
        ml.get_key(LLM_KV_FULL_ATTENTION_INTERVAL, full_attn_interval, false);
        for (uint32_t i = 0; i < hparams.n_layer_all; ++i) {
            hparams.is_recr_impl[i] = (i < hparams.n_layer()) && ((i + 1) % full_attn_interval != 0);
        }
    }

    switch (hparams.n_layer()) {
        case 24: type = hparams.n_embd == 1024 ? LLM_TYPE_0_8B : LLM_TYPE_2B; break;
        case 32: type = hparams.n_embd == 2560 ? LLM_TYPE_4B : LLM_TYPE_9B; break;
        case 64: type = LLM_TYPE_27B; break;
        default: type = LLM_TYPE_UNKNOWN;
    }
}

void llama_model_qwen35::load_arch_tensors(llama_model_loader & ml) {
    LLAMA_LOAD_LOCALS;

    const bool mtp_only = (hparams.n_layer_nextn > 0) && (ml.get_weight("blk.0.attn_norm.weight") == nullptr);
    const int trunk_flags = mtp_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 }, TENSOR_NOT_REQUIRED);

    // if output is NULL, init from the input tok embed
    if (output == NULL) {
        output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), { n_embd, n_vocab }, TENSOR_DUPLICATED);
    }

    // Optional sequence-classification projection. The generic RANK pooling
    // graph applies this weight to the selected final non-padding token.
    cls_out = create_tensor(tn(LLM_TENSOR_CLS_OUT, "weight"), { n_embd, hparams.n_cls_out }, TENSOR_NOT_REQUIRED);
    // Optional early-exit head (same shape). Used when SCAMGUARDIAN_EARLY_EXIT_LAYERS is set.
    cls_out_early = create_tensor(tn(LLM_TENSOR_CLS_OUT_EARLY, "weight"), { n_embd, hparams.n_cls_out }, TENSOR_NOT_REQUIRED);
    // Optional joint evidence heads. They are consumed only by the opt-in
    // ScamGuardian span report path; legacy technique-only GGUFs omit them.
    cls_span_start = create_tensor(tn(LLM_TENSOR_CLS_SPAN_START, "weight"), { n_embd, hparams.n_cls_out }, TENSOR_NOT_REQUIRED);
    cls_span_start_b = create_tensor(tn(LLM_TENSOR_CLS_SPAN_START, "bias"), { hparams.n_cls_out }, TENSOR_NOT_REQUIRED);
    cls_span_end = create_tensor(tn(LLM_TENSOR_CLS_SPAN_END, "weight"), { n_embd, hparams.n_cls_out }, TENSOR_NOT_REQUIRED);
    cls_span_end_b = create_tensor(tn(LLM_TENSOR_CLS_SPAN_END, "bias"), { hparams.n_cls_out }, TENSOR_NOT_REQUIRED);
    cls_span_pair = create_tensor(tn(LLM_TENSOR_CLS_SPAN_PAIR, "weight"), { 2 * n_embd, hparams.n_cls_out }, TENSOR_NOT_REQUIRED);
    cls_span_pair_b = create_tensor(tn(LLM_TENSOR_CLS_SPAN_PAIR, "bias"), { hparams.n_cls_out }, TENSOR_NOT_REQUIRED);

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

        // Calculate dimensions from hyperparameters
        const int64_t head_k_dim = hparams.ssm_d_state;
        const int64_t head_v_dim = hparams.ssm_d_state;
        const int64_t n_k_heads  = hparams.ssm_n_group;
        const int64_t n_v_heads  = hparams.ssm_dt_rank;
        const int64_t key_dim    = head_k_dim * n_k_heads;
        const int64_t value_dim  = head_v_dim * n_v_heads;
        const int64_t conv_dim   = key_dim * 2 + value_dim;

        layer.attn_norm      = create_tensor(tn(LLM_TENSOR_ATTN_NORM,      "weight", il), { n_embd }, flags);
        layer.attn_post_norm = create_tensor(tn(LLM_TENSOR_ATTN_POST_NORM, "weight", il), { n_embd }, flags);

        if (!hparams.is_recr(il)) {
            // Attention layers
            create_tensor_qkv(layer, il, n_embd, n_embd_head_k * n_head * 2, n_embd_k_gqa, n_embd_v_gqa, flags);
            layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", il), { n_embd_head_k * n_head, n_embd }, flags);

            // Q/K normalization for attention layers
            layer.attn_q_norm = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "weight", il), { n_embd_head_k }, flags);
            layer.attn_k_norm = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "weight", il), { n_embd_head_k }, flags);
        } else {
            // Linear attention (gated delta net) specific tensors
            // Create tensors with calculated dimensions
            layer.wqkv           = create_tensor(tn(LLM_TENSOR_ATTN_QKV,       "weight", il), { n_embd, key_dim * 2 + value_dim }, TENSOR_NOT_REQUIRED);
            layer.wqkv_gate      = create_tensor(tn(LLM_TENSOR_ATTN_GATE,      "weight", il), { n_embd, value_dim }, TENSOR_NOT_REQUIRED);
            layer.ssm_conv1d     = create_tensor(tn(LLM_TENSOR_SSM_CONV1D,     "weight", il), { hparams.ssm_d_conv, conv_dim }, flags);
            layer.ssm_dt         = create_tensor(tn(LLM_TENSOR_SSM_DT,         "bias",   il), { hparams.ssm_dt_rank }, flags);
            layer.ssm_a          = create_tensor(tn(LLM_TENSOR_SSM_A_NOSCAN,             il), { hparams.ssm_dt_rank }, flags);
            layer.ssm_beta       = create_tensor(tn(LLM_TENSOR_SSM_BETA,       "weight", il), { n_embd, n_v_heads }, flags);
            layer.ssm_alpha      = create_tensor(tn(LLM_TENSOR_SSM_ALPHA,      "weight", il), { n_embd, n_v_heads }, flags);
            layer.ssm_norm       = create_tensor(tn(LLM_TENSOR_SSM_NORM,       "weight", il), { head_v_dim }, flags);
            layer.ssm_out        = create_tensor(tn(LLM_TENSOR_SSM_OUT,        "weight", il), { value_dim, n_embd }, flags);
        }

        layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", il), {n_embd,   n_ff}, flags);
        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", il), {  n_ff, n_embd}, flags);
        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", il), {n_embd,   n_ff}, flags);
    };

    auto load_block_mtp = [&](int il) {
        auto & layer = layers[il];

        // MTP block looks like a full-attention Qwen3.5 decoder block.
        layer.attn_norm      = create_tensor(tn(LLM_TENSOR_ATTN_NORM,      "weight", il), { n_embd }, 0);
        layer.attn_post_norm = create_tensor(tn(LLM_TENSOR_ATTN_POST_NORM, "weight", il), { n_embd }, 0);

        create_tensor_qkv(layer, il, n_embd, n_embd_head_k * n_head * 2, n_embd_k_gqa, n_embd_v_gqa, 0);
        layer.wo          = create_tensor(tn(LLM_TENSOR_ATTN_OUT,    "weight", il), { n_embd_head_k * n_head, n_embd }, 0);
        layer.attn_q_norm = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "weight", il), { n_embd_head_k }, 0);
        layer.attn_k_norm = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "weight", il), { n_embd_head_k }, 0);

        layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", il), {n_embd,   n_ff}, 0);
        layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", il), {  n_ff, n_embd}, 0);
        layer.ffn_up   = create_tensor(tn(LLM_TENSOR_FFN_UP,   "weight", il), {n_embd,   n_ff}, 0);

        // NextN-specific tensors that define the MTP block.
        layer.nextn.eh_proj          = create_tensor(tn(LLM_TENSOR_NEXTN_EH_PROJ,          "weight", il), { 2 * n_embd, n_embd }, 0);
        layer.nextn.enorm            = create_tensor(tn(LLM_TENSOR_NEXTN_ENORM,            "weight", il), { n_embd },              0);
        layer.nextn.hnorm            = create_tensor(tn(LLM_TENSOR_NEXTN_HNORM,            "weight", il), { n_embd },              0);
        layer.nextn.embed_tokens     = create_tensor(tn(LLM_TENSOR_NEXTN_EMBED_TOKENS,     "weight", il), { n_embd, n_vocab },     TENSOR_NOT_REQUIRED);
        layer.nextn.shared_head_head = create_tensor(tn(LLM_TENSOR_NEXTN_SHARED_HEAD_HEAD, "weight", il), { n_embd, n_vocab },     TENSOR_NOT_REQUIRED);
        layer.nextn.shared_head_norm = create_tensor(tn(LLM_TENSOR_NEXTN_SHARED_HEAD_NORM, "weight", il), { n_embd },              TENSOR_NOT_REQUIRED);
    };

    for (int i = 0; i < n_layer; ++i) {
        load_block_trunk(i, trunk_flags);
    }
    for (int i = n_layer; i < n_layer_all; ++i) {
        load_block_mtp(i);
    }
}

std::unique_ptr<llm_graph_context> llama_model_qwen35::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_qwen35::graph::graph(const llama_model & model, const llm_graph_params & params) :
    llm_build_delta_net_base(params), model(model) {
    const int64_t n_embd_head = hparams.n_embd_head_v();

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

    int sections[4];
    std::copy(std::begin(hparams.rope_sections), std::begin(hparams.rope_sections) + 4, sections);

    ggml_tensor * cur;
    ggml_tensor * inpL;

    const int layer_start = cparams.scamguardian_layer_start;
    if (layer_start > 0) {
        GGML_ASSERT(hparams.n_embd_inp() == hparams.n_embd);

        auto inp = std::make_unique<llm_graph_input_embd>(
            hparams.n_embd_inp());
        inp->embd = ggml_new_tensor_2d(
            ctx0,
            GGML_TYPE_F32,
            hparams.n_embd_inp(),
            ubatch.n_tokens);
        cb(inp->embd, "continuation_input", -1);
        ggml_set_input(inp->embd);
        inpL = inp->embd;
        res->t_inp_embd = inpL;
        res->add_input(std::move(inp));
        ggml_build_forward_expand(gf, inpL);
    } else {
        inpL = build_inp_embd(model.tok_embd);
    }

    cb(inpL, "model.input_embed", -1);

    auto * inp = build_inp_mem_hybrid();

    // Experimental ScamGuardian classifier-only no-KV path.
    // Disabled by default so existing runtime behavior remains unchanged.
    const char * no_kv_env = std::getenv("SCAMGUARDIAN_QWEN35_NO_KV");

    const bool classifier_no_kv =
        no_kv_env != nullptr &&
        no_kv_env[0] == '1' &&
        no_kv_env[1] == '\0' &&
        cparams.embeddings &&
        pooling_type == LLAMA_POOLING_TYPE_RANK &&
        model.cls_out != nullptr &&
        hparams.n_cls_out > 1;

    const bool specialized_tail = q35_specialize_level() >= 2 && classifier_no_kv;
    if (specialized_tail) {
        // Ported: the embedding row count is a model-identity guard only. The specialized
        // tail keeps the last four token rows and skips inp_out_ids; neither depends on
        // the vocabulary size. Depth, width and head width are still enforced below.
        GGML_ASSERT(n_layer == 4 && hparams.n_embd == 1024 && model.tok_embd->ne[1] > 0);
        GGML_ASSERT(hparams.n_cls_out == 14 && ubatch.n_seqs == 1 && n_tokens <= 256);
        GGML_ASSERT(!hparams.is_recr(3) && layer_start == 0);
    }
    // Retain the final four rows to preserve the native Q8 GEMM tile route.
    const int64_t tail_rows = std::min<int64_t>(4, n_tokens);

    llm_graph_input_attn_no_cache * inp_attn_no_cache =
        classifier_no_kv ? build_attn_inp_no_cache() : nullptr;

    inp->set_attn_input_enabled(!classifier_no_kv);

    ggml_tensor * inp_pos     = build_inp_pos();
    ggml_tensor * inp_out_ids = specialized_tail ? nullptr : build_inp_out_ids();

    const char * debug_stage = std::getenv("SCAMGUARDIAN_LAYER_DUMP_STAGE");
    if (debug_stage == nullptr) {
        debug_stage = "layer_input";
    }
    auto stage_is = [debug_stage](const char * name) {
        return std::strcmp(debug_stage, name) == 0;
    };
    auto set_debug_output = [&](const char * stage, ggml_tensor * tensor, int il) {
        if (!stage_is(stage)) {
            return;
        }

        // Debug intermediates are otherwise eligible for scheduler buffer reuse.
        // Copy each selected layer result to an independent graph output so a
        // later layer cannot overwrite the value before the host extracts it.
        ggml_tensor * dst = ggml_new_tensor_4d(
            ctx0,
            GGML_TYPE_F32,
            tensor->ne[0],
            tensor->ne[1],
            tensor->ne[2],
            tensor->ne[3]);
        ggml_format_name(dst, "scamguardian_%s_dst_%d", stage, il);

        ggml_tensor * copy = ggml_cpy(ctx0, tensor, dst);
        ggml_format_name(copy, "scamguardian_%s_copy_%d", stage, il);
        ggml_build_forward_expand(gf, copy);
        res->t_layer_inp[il] = copy;
    };

    // Forced extract_only / early-exit for classifier embeddings.
    // Caps the decoder loop while keeping the full GGUF resident (suffix weights unused).
    // Prefers cls_out_early when present; otherwise falls back to final cls_out on early hidden.
    // Env: SCAMGUARDIAN_EARLY_EXIT_LAYERS=<N> or LAYERS_FILE with a single int.
    // When SCAMGUARDIAN_EARLY_EXIT_DYNAMIC=1, graph reuse is disabled (see llama-graph.cpp).
    int n_layer_run = cparams.scamguardian_layer_end > 0
        ? cparams.scamguardian_layer_end
        : n_layer;
    if (cparams.scamguardian_layer_end == 0 &&
        cparams.embeddings && model.cls_out != nullptr) {
        const int early_layers = scamguardian_early_exit_layers(n_layer);
        if (early_layers > 0) {
            n_layer_run = early_layers;
        }
    }
    GGML_ASSERT(layer_start >= 0 && layer_start < n_layer_run);
    const int continuation_boundary =
        scamguardian_continuation_boundary(n_layer);
    // MTP/NextN layers are loaded as extra decoder blocks but not executed in the main pass.
    for (int il = layer_start; il < n_layer_run; ++il) {
        if (stage_is("layer_input")) {
            res->t_layer_inp[il] = inpL;
        }

        ggml_tensor * inpSA = inpL;

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

        set_debug_output("attn_norm", cur, il);

        ggml_build_forward_expand(gf, cur);

        // Determine layer type and build appropriate attention mechanism
        if (hparams.is_recr(il)) {
            // Linear attention layer (gated delta net).
            // The recurrent state remains in llama_memory_hybrid.
            cur = build_layer_attn_linear(inp->get_recr(), cur, il);
        } else if (classifier_no_kv) {
            // Single-shot classifier path: consume current Q/K/V directly.
            cur = build_layer_attn(
                inp_attn_no_cache,
                cur,
                inp_pos,
                sections,
                il);
        } else {
            // Existing KV-backed path.
            cur = build_layer_attn(
                inp->get_attn(),
                cur,
                inp_pos,
                sections,
                il);
        }

        set_debug_output("attn_out", cur, il);

        if (il == n_layer_run - 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);
        }

        if (specialized_tail && il == n_layer_run - 1) {
            cur = ggml_view_2d(ctx0, cur, n_embd, tail_rows, cur->nb[1],
                              (cur->ne[1] - tail_rows) * cur->nb[1]);
            inpSA = ggml_view_2d(ctx0, inpSA, n_embd, tail_rows, inpSA->nb[1],
                                (inpSA->ne[1] - tail_rows) * inpSA->nb[1]);
        }

        // Residual connection
        cur = ggml_add(ctx0, cur, inpSA);
        cb(cur, "attn_residual", il);

        set_debug_output("attn_residual", cur, il);

        // Save the tensor before post-attention norm for residual connection
        ggml_tensor * ffn_residual = cur;

        // Post-attention norm
        ggml_tensor * attn_post_norm = build_norm(cur, model.layers[il].attn_post_norm, nullptr, LLM_NORM_RMS, il);
        cb(attn_post_norm, "attn_post_norm", il);

        set_debug_output("attn_post_norm", attn_post_norm, il);

        // Dense FFN layer - without residual connection
        cur = build_layer_ffn(attn_post_norm, il);
        cb(cur, "ffn_out", il);

        set_debug_output("ffn_out", cur, il);

        // Residual connection for FFN - add to the tensor from before post_attention_layernorm
        cur = ggml_add(ctx0, cur, ffn_residual);
        cb(cur, "post_ffn", il);

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

        if (continuation_boundary == il + 1) {
            cur = ggml_cont(ctx0, cur);
            cb(cur, "continuation_boundary", il);
            ggml_build_forward_expand(gf, cur);
        }

        if (stage_is("layer_out")) {
            res->t_layer_inp[il] = cur;
        }

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

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

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

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

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

    // A classifier checkpoint does not need the 248K-token vocabulary
    // projection. RANK pooling consumes result_norm and applies cls_out.
    if (cparams.embeddings && pooling_type == LLAMA_POOLING_TYPE_RANK && model.cls_out != nullptr) {
        ggml_build_forward_expand(gf, cur);
        return;
    }

    // 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);
}

std::pair<ggml_tensor *, ggml_tensor *> llama_model_qwen35::graph::build_qkvz(
                ggml_tensor * input,
                        int   il) {
    const int64_t n_seqs       = ubatch.n_seqs;
    const int64_t n_seq_tokens = ubatch.n_seq_tokens;

    ggml_tensor * qkv_mixed = build_lora_mm(model.layers[il].wqkv, input, model.layers[il].wqkv_s);
    qkv_mixed = ggml_reshape_3d(ctx0, qkv_mixed, qkv_mixed->ne[0], n_seq_tokens, n_seqs);
    cb(qkv_mixed, "linear_attn_qkv_mixed", il);

    ggml_tensor * z = build_lora_mm(model.layers[il].wqkv_gate, input, model.layers[il].wqkv_gate_s);
    cb(z, "z", il);

    return { qkv_mixed, z };
}

ggml_tensor * llama_model_qwen35::graph::build_norm_gated(
        ggml_tensor * input,
        ggml_tensor * weights,
        ggml_tensor * gate,
        int           layer) {
    ggml_tensor * normalized = build_norm(input, weights, nullptr, LLM_NORM_RMS, layer);
    ggml_tensor * gated_silu = ggml_silu(ctx0, gate);

    return ggml_mul(ctx0, normalized, gated_silu);
}

ggml_tensor * llama_model_qwen35::graph::build_layer_attn(
        llm_graph_input_attn_kv * inp,
        ggml_tensor *             cur,
        ggml_tensor *             inp_pos,
        int *                     sections,
        int                       il) {
    const int64_t n_embd_head = hparams.n_embd_head_v();
    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k());

    // Order: joint QG projection, QG split, Q norm, KV projection, K norm, RoPE, attention

    // Qwen3Next uses a single Q projection that outputs query + gate
    ggml_tensor * Qcur_full = build_lora_mm(model.layers[il].wq, cur, model.layers[il].wq_s); // [ (n_embd_head * 2) * n_head, n_tokens ]
    cb(Qcur_full, "Qcur_full", il);

    ggml_tensor * Qcur = ggml_view_3d(ctx0, Qcur_full, n_embd_head, n_head, n_tokens,
        ggml_element_size(Qcur_full) * n_embd_head * 2,
        ggml_element_size(Qcur_full) * n_embd_head * 2 * n_head, 0);
    cb(Qcur, "Qcur_reshaped", il);

    // Apply Q normalization
    Qcur = build_norm(Qcur, model.layers[il].attn_q_norm, nullptr, LLM_NORM_RMS, il);
    cb(Qcur, "Qcur_normed", il);

    ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur, model.layers[il].wk_s);
    cb(Kcur, "Kcur", il);

    ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur, model.layers[il].wv_s);
    cb(Vcur, "Vcur", il);

    // Apply K normalization
    Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
    Kcur = build_norm(Kcur, model.layers[il].attn_k_norm, nullptr, LLM_NORM_RMS, il);
    cb(Kcur, "Kcur_normed", il);

    ggml_tensor * gate = ggml_view_3d(ctx0, Qcur_full, n_embd_head, n_head, n_tokens,
        ggml_element_size(Qcur_full) * n_embd_head * 2,
        ggml_element_size(Qcur_full) * n_embd_head * 2 * n_head,
        ggml_element_size(Qcur_full) * n_embd_head);
    gate = ggml_cont_2d(ctx0, gate, n_embd_head * n_head, n_tokens);
    cb(gate, "gate_reshaped", il);

    Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);

    // Apply MRoPE
    Qcur = ggml_rope_multi(
            ctx0, Qcur, inp_pos, nullptr,
            n_rot, sections, rope_type, n_ctx_orig, freq_base, freq_scale,
            ext_factor, attn_factor, beta_fast, beta_slow
            );

    Kcur = ggml_rope_multi(
            ctx0, Kcur, inp_pos, nullptr,
            n_rot, sections, 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);

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

    cur = build_attn(inp,
                nullptr, nullptr, nullptr,
                Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, kq_scale, il);
    cb(cur, "attn_pregate", il);

    ggml_tensor * gate_sigmoid = ggml_sigmoid(ctx0, gate);
    cb(gate_sigmoid, "gate_sigmoid", il);

    cur = ggml_mul(ctx0, cur, gate_sigmoid);
    cb(cur, "attn_gated", il);

    cur = build_lora_mm(model.layers[il].wo, cur, model.layers[il].wo_s);
    cb(cur, "attn_output", il);

    return cur;
}

ggml_tensor * llama_model_qwen35::graph::build_layer_attn(
        llm_graph_input_attn_no_cache * inp,
        ggml_tensor *             cur,
        ggml_tensor *             inp_pos,
        int *                     sections,
        int                       il) {
    const int64_t n_embd_head = hparams.n_embd_head_v();
    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k());

    const int64_t q_tokens = q35_specialize_level() >= 3 && il == 3
        ? std::min<int64_t>(4, n_tokens) : n_tokens;
    ggml_tensor * q_input = cur;
    ggml_tensor * q_pos = inp_pos;
    if (q_tokens < n_tokens) {
        q_input = ggml_view_2d(ctx0, cur, cur->ne[0], q_tokens, cur->nb[1], (n_tokens-q_tokens)*cur->nb[1]);
        const int64_t axes = inp_pos->ne[0] / n_tokens;
        q_pos = ggml_cont(ctx0, ggml_view_2d(ctx0, inp_pos, q_tokens, axes,
                          n_tokens * inp_pos->nb[0], (n_tokens-q_tokens)*inp_pos->nb[0]));
        q_pos = ggml_reshape_1d(ctx0, q_pos, q_tokens * axes);
    }

    // Order: joint QG projection, QG split, Q norm, KV projection, K norm, RoPE, attention

    // Qwen3Next uses a single Q projection that outputs query + gate
    ggml_tensor * Qcur_full = build_lora_mm(model.layers[il].wq, q_input, model.layers[il].wq_s); // [ (n_embd_head * 2) * n_head, n_tokens ]
    cb(Qcur_full, "Qcur_full", il);

    ggml_tensor * Qcur = ggml_view_3d(ctx0, Qcur_full, n_embd_head, n_head, q_tokens,
        ggml_element_size(Qcur_full) * n_embd_head * 2,
        ggml_element_size(Qcur_full) * n_embd_head * 2 * n_head, 0);
    cb(Qcur, "Qcur_reshaped", il);

    // Apply Q normalization
    Qcur = build_norm(Qcur, model.layers[il].attn_q_norm, nullptr, LLM_NORM_RMS, il);
    cb(Qcur, "Qcur_normed", il);

    ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur, model.layers[il].wk_s);
    cb(Kcur, "Kcur", il);

    ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur, model.layers[il].wv_s);
    cb(Vcur, "Vcur", il);

    // Apply K normalization
    Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
    Kcur = build_norm(Kcur, model.layers[il].attn_k_norm, nullptr, LLM_NORM_RMS, il);
    cb(Kcur, "Kcur_normed", il);

    ggml_tensor * gate = ggml_view_3d(ctx0, Qcur_full, n_embd_head, n_head, q_tokens,
        ggml_element_size(Qcur_full) * n_embd_head * 2,
        ggml_element_size(Qcur_full) * n_embd_head * 2 * n_head,
        ggml_element_size(Qcur_full) * n_embd_head);
    gate = ggml_cont_2d(ctx0, gate, n_embd_head * n_head, q_tokens);
    cb(gate, "gate_reshaped", il);

    Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);

    // Apply MRoPE
    Qcur = ggml_rope_multi(
            ctx0, Qcur, q_pos, nullptr,
            n_rot, sections, rope_type, n_ctx_orig, freq_base, freq_scale,
            ext_factor, attn_factor, beta_fast, beta_slow
            );

    Kcur = ggml_rope_multi(
            ctx0, Kcur, inp_pos, nullptr,
            n_rot, sections, 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);

    // Attention computation without persistent K/V storage.
    const float kq_scale = hparams.f_attention_scale == 0.0f ? 1.0f / sqrtf(float(n_embd_head)) : hparams.f_attention_scale;

    cur = build_attn(inp,
                nullptr, nullptr, nullptr,
                Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, kq_scale, il);
    cb(cur, "attn_pregate", il);

    ggml_tensor * gate_sigmoid = ggml_sigmoid(ctx0, gate);
    cb(gate_sigmoid, "gate_sigmoid", il);

    cur = ggml_mul(ctx0, cur, gate_sigmoid);
    cb(cur, "attn_gated", il);

    cur = build_lora_mm(model.layers[il].wo, cur, model.layers[il].wo_s);
    cb(cur, "attn_output", il);

    return cur;
}

ggml_tensor * llama_model_qwen35::graph::build_layer_attn_linear(
        llm_graph_input_rs * inp,
        ggml_tensor *        cur,
        int                  il) {
    const auto * mctx_cur = inp->mctx;

    const int64_t d_inner      = hparams.ssm_d_inner;
    const int64_t n_seqs       = ubatch.n_seqs;
    const int64_t head_k_dim   = hparams.ssm_d_state;
    const int64_t num_k_heads  = hparams.ssm_n_group;
    const int64_t num_v_heads  = hparams.ssm_dt_rank;
    const int64_t head_v_dim   = d_inner / num_v_heads;
    const int64_t n_seq_tokens = ubatch.n_seq_tokens;

    GGML_ASSERT(n_seqs != 0);
    GGML_ASSERT(ubatch.equal_seqs());
    GGML_ASSERT(ubatch.n_tokens == n_seq_tokens * n_seqs);

    // Input projections
    auto qkvz = build_qkvz(cur, il);
    ggml_tensor * qkv_mixed = qkvz.first;
    ggml_tensor * z         = qkvz.second;

    ggml_tensor * beta = build_lora_mm(model.layers[il].ssm_beta, cur, model.layers[il].ssm_beta_s);
    beta = ggml_reshape_4d(ctx0, beta, 1, num_v_heads, n_seq_tokens, n_seqs);
    cb(beta, "beta", il);

    beta = ggml_sigmoid(ctx0, beta);
    cb(beta, "beta_sigmoid", il);

    ggml_tensor * alpha = build_lora_mm(model.layers[il].ssm_alpha, cur, model.layers[il].ssm_alpha_s);
    alpha = ggml_reshape_3d(ctx0, alpha, num_v_heads, n_seq_tokens, n_seqs);
    cb(alpha, "alpha", il);

    ggml_tensor * alpha_biased   = ggml_add(ctx0, alpha, model.layers[il].ssm_dt);
    ggml_tensor * alpha_softplus = ggml_softplus(ctx0, alpha_biased);
    cb(alpha_softplus, "a_softplus", il);

    ggml_tensor * gate = ggml_mul(ctx0, alpha_softplus, model.layers[il].ssm_a);  // -A_log.exp() * softplus
    cb(gate, "gate", il);

    gate = ggml_reshape_4d(ctx0, gate, 1, num_v_heads, n_seq_tokens, n_seqs);

    ggml_tensor * conv_states_all = mctx_cur->get_r_l(il);
    ggml_tensor * ssm_states_all  = mctx_cur->get_s_l(il);

    ggml_tensor * conv_kernel      = model.layers[il].ssm_conv1d;
    const int64_t conv_kernel_size = conv_kernel->ne[0];
    const int64_t conv_channels    = d_inner + 2 * hparams.ssm_n_group * hparams.ssm_d_state;

    ggml_tensor * conv_input = build_conv_state(inp, conv_states_all, qkv_mixed, conv_kernel_size, conv_channels, il);

    ggml_tensor * state = build_rs(inp, ssm_states_all, hparams.n_embd_s(), n_seqs);
    state = ggml_reshape_4d(ctx0, state, head_v_dim, head_v_dim, num_v_heads, n_seqs);
    cb(state, "state_predelta", il);

    ggml_tensor * conv_output_proper = ggml_ssm_conv(ctx0, conv_input, conv_kernel);
    cb(conv_output_proper, "conv_output_raw", il);

    ggml_tensor * conv_output_silu = ggml_silu(ctx0, conv_output_proper);
    cb(conv_output_silu, "conv_output_silu", il);

    ggml_tensor * conv_qkv_mix = conv_output_silu;

    // Calculate the total conv dimension
    int64_t qkv_dim = head_k_dim * num_k_heads * 2 + head_v_dim * num_v_heads;
    int64_t nb1_qkv = ggml_row_size(conv_qkv_mix->type, qkv_dim);

    // Extract the convolved Q, K, V from conv_output
    ggml_tensor * q_conv = ggml_view_4d(ctx0, conv_qkv_mix, head_k_dim, num_k_heads, n_seq_tokens, n_seqs,
            ggml_row_size(conv_qkv_mix->type, head_k_dim),
            nb1_qkv,
            nb1_qkv * n_seq_tokens,
            0);

    ggml_tensor * k_conv = ggml_view_4d(ctx0, conv_qkv_mix, head_k_dim, num_k_heads, n_seq_tokens, n_seqs,
            ggml_row_size(conv_qkv_mix->type, head_k_dim),
            nb1_qkv,
            nb1_qkv * n_seq_tokens,
            head_k_dim * num_k_heads * ggml_element_size(conv_qkv_mix));

    ggml_tensor * v_conv = ggml_view_4d(ctx0, conv_qkv_mix, head_v_dim, num_v_heads, n_seq_tokens, n_seqs,
            ggml_row_size(conv_qkv_mix->type, head_v_dim),
            nb1_qkv,
            nb1_qkv * n_seq_tokens,
            ggml_row_size(conv_qkv_mix->type, 2 * head_k_dim * num_k_heads));

    cb(q_conv, "q_conv", il);
    cb(k_conv, "k_conv", il);
    cb(v_conv, "v_conv", il);

    const float eps_norm = hparams.f_norm_rms_eps;

    q_conv = ggml_l2_norm(ctx0, q_conv, eps_norm);
    k_conv = ggml_l2_norm(ctx0, k_conv, eps_norm);

    //q_conv = ggml_cont_4d(ctx0, q_conv, head_k_dim, num_k_heads, n_seq_tokens, n_seqs);
    //k_conv = ggml_cont_4d(ctx0, k_conv, head_k_dim, num_k_heads, n_seq_tokens, n_seqs);
    //v_conv = ggml_cont_4d(ctx0, v_conv, head_v_dim, num_v_heads, n_seq_tokens, n_seqs);

    // if head keys and value keys are different, repeat to force tensors into matching shapes
    // note: need explicit repeat only if we are not using the fused GDN.
    if (num_k_heads != num_v_heads && (!cparams.fused_gdn_ar || !cparams.fused_gdn_ch)) {
        GGML_ASSERT(num_v_heads % num_k_heads == 0);
        q_conv = ggml_repeat_4d(ctx0, q_conv, head_k_dim, num_v_heads, n_seq_tokens, n_seqs);
        k_conv = ggml_repeat_4d(ctx0, k_conv, head_k_dim, num_v_heads, n_seq_tokens, n_seqs);
    }

    cb(q_conv, "q_conv_predelta", il);
    cb(k_conv, "k_conv_predelta", il);
    cb(v_conv, "v_conv_predelta", il);

    ggml_tensor * output = build_recurrent_attn(inp, ssm_states_all, q_conv, k_conv, v_conv, gate, beta, state, il);

    // z: [head_dim, n_heads, n_tokens, n_seqs] -> [n_heads * n_tokens * n_seqs, head_dim]
    ggml_tensor * z_2d = ggml_reshape_4d(ctx0, z, head_v_dim, num_v_heads, n_seq_tokens, n_seqs);

    // Apply gated normalization: self.norm(core_attn_out, z)
    ggml_tensor * attn_out_norm = build_norm_gated(output, model.layers[il].ssm_norm, z_2d, il);

    // Final reshape: [head_dim, n_heads, n_tokens, n_seqs] -> [n_tokens, n_seqs, n_heads * head_dim]
    ggml_tensor * final_output = ggml_reshape_3d(ctx0, attn_out_norm, head_v_dim * num_v_heads, n_seq_tokens, n_seqs);
    cb(final_output, "final_output", il);

    // Output projection
    cur = build_lora_mm(model.layers[il].ssm_out, final_output, model.layers[il].ssm_out_s);
    cb(cur, "linear_attn_out", il);

    // Reshape back to original dimensions
    cur = ggml_reshape_2d(ctx0, cur, n_embd, n_seq_tokens * n_seqs);

    return cur;
}

ggml_tensor * llama_model_qwen35::graph::build_layer_ffn(ggml_tensor * cur, const int il) {
    // Qwen3.5 does not use MoE FFN
    GGML_ASSERT(model.layers[il].ffn_gate_inp == nullptr);

    cur = build_ffn(cur,
        model.layers[il].ffn_up, NULL, model.layers[il].ffn_up_s,
        model.layers[il].ffn_gate, NULL, model.layers[il].ffn_gate_s,
        model.layers[il].ffn_down, NULL, model.layers[il].ffn_down_s,
        NULL,
        LLM_FFN_SILU, LLM_FFN_PAR, il);
    cb(cur, "ffn_out", il);

    return cur;
}

// LLM_GRAPH_TYPE_DECODER_MTP draft head for Qwen3.5/3.6 dense series
llama_model_qwen35::graph_mtp::graph_mtp(const llama_model & model, const llm_graph_params & params)
    : llm_graph_context(params) {
    GGML_ASSERT(hparams.n_layer_nextn > 0 && "QWEN35 MTP requires n_layer_nextn > 0");
    GGML_ASSERT(hparams.n_layer_nextn == 1 && "QWEN35 MTP currently only supports a single MTP block");

    const int64_t n_embd_head = hparams.n_embd_head_v();
    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k());

    // hparams.n_layer includes both main model layers and MTP layers. The MTP
    // layer is stored immediately after the main layers in model.layers[].
    const int il = hparams.n_layer();
    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");

    int sections[4];
    std::copy(std::begin(hparams.rope_sections), std::begin(hparams.rope_sections) + 4, sections);

    // TODO: extract in a common llm_graph_context::build_inp_embd_h()
    auto inp = std::make_unique<llm_graph_input_embd_h>(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_inp(), n_tokens);
    ggml_set_input(inp->embd);

    // TODO: make static using `ggml_build_forward_select()`
    //       see llm_graph_context::build_inp_embd() for reference
    ggml_tensor * tok_embd;
    if (ubatch.token) {
        ggml_tensor * tok_embd_w = layer.nextn.embed_tokens ? layer.nextn.embed_tokens : model.tok_embd;

        tok_embd = ggml_get_rows(ctx0, tok_embd_w, inp->tokens);
    } else {
        tok_embd = inp->embd;
    }
    cb(tok_embd, "mtp_tok_embd", il);

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

    ggml_tensor * h_embd = inp->h;

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

    ggml_tensor * inp_pos     = build_inp_pos();
    ggml_tensor * inp_out_ids = build_inp_out_ids();

    auto * inp_attn = build_attn_inp_kv();

    ggml_tensor * h_norm = build_norm(h_embd, 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, layer.nextn.eh_proj_s);
    cb(cur, "mtp_eh_proj", il);

    ggml_tensor * inpSA = cur;

    cur = build_norm(cur, layer.attn_norm, nullptr, LLM_NORM_RMS, il);
    cb(cur, "mtp_attn_norm", il);

    ggml_tensor * Qcur_full = build_lora_mm(layer.wq, cur, layer.wq_s);
    cb(Qcur_full, "mtp_Qcur_full", il);

    ggml_tensor * Qcur = ggml_view_3d(ctx0, Qcur_full,
            n_embd_head, n_head, n_tokens,
            ggml_element_size(Qcur_full) * n_embd_head * 2,
            ggml_element_size(Qcur_full) * n_embd_head * 2 * n_head,
            0);
    Qcur = build_norm(Qcur, layer.attn_q_norm, nullptr, LLM_NORM_RMS, il);
    cb(Qcur, "mtp_Qcur_normed", il);

    ggml_tensor * gate = ggml_view_3d(ctx0, Qcur_full,
            n_embd_head, n_head, n_tokens,
            ggml_element_size(Qcur_full) * n_embd_head * 2,
            ggml_element_size(Qcur_full) * n_embd_head * 2 * n_head,
            ggml_element_size(Qcur_full) * n_embd_head);
    gate = ggml_cont_2d(ctx0, gate, n_embd_head * n_head, n_tokens);
    cb(gate, "mtp_gate", il);

    ggml_tensor * Kcur = build_lora_mm(layer.wk, cur, layer.wk_s);
    Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
    Kcur = build_norm(Kcur, layer.attn_k_norm, nullptr, LLM_NORM_RMS, il);
    cb(Kcur, "mtp_Kcur_normed", il);

    ggml_tensor * Vcur = build_lora_mm(layer.wv, cur, layer.wv_s);
    Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
    cb(Vcur, "mtp_Vcur", il);

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

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

    cur = build_attn(inp_attn,
            nullptr, nullptr, nullptr,
            Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, kq_scale, il);
    cb(cur, "mtp_attn_pregate", il);

    cur = ggml_mul(ctx0, cur, ggml_sigmoid(ctx0, gate));
    cur = build_lora_mm(layer.wo, cur, layer.wo_s);
    cb(cur, "mtp_attn_out", il);

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

    ggml_tensor * ffn_residual = cur;
    cur = build_norm(cur, layer.attn_post_norm, nullptr, LLM_NORM_RMS, il);
    cb(cur, "mtp_attn_post_norm", il);

    cur = build_ffn(cur,
            layer.ffn_up,   nullptr, layer.ffn_up_s,
            layer.ffn_gate, nullptr, layer.ffn_gate_s,
            layer.ffn_down, nullptr, layer.ffn_down_s,
            nullptr,
            LLM_FFN_SILU, LLM_FFN_PAR, il);
    cb(cur, "mtp_ffn_out", il);

    cur = ggml_add(ctx0, cur, ffn_residual);
    cb(cur, "mtp_post_ffn", il);

    ggml_tensor * head_norm_w = layer.nextn.shared_head_norm
            ? layer.nextn.shared_head_norm
            : model.output_norm;
    GGML_ASSERT(head_norm_w && "QWEN35 MTP: missing both nextn.shared_head_norm and output_norm");
    cur = build_norm(cur, head_norm_w, nullptr, LLM_NORM_RMS, -1);

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

    cur = ggml_get_rows(ctx0, cur, inp_out_ids);
    cb(cur, "mtp_shared_head_norm", -1);

    ggml_tensor * head_w = layer.nextn.shared_head_head ? layer.nextn.shared_head_head : model.output;
    ggml_tensor * head_s = layer.nextn.shared_head_head ? layer.nextn.shared_head_head_s : model.output_s;
    GGML_ASSERT(head_w && "QWEN35 MTP: missing LM head (nextn.shared_head_head or model.output)");
    cur = build_lora_mm(head_w, cur, head_s);
    cb(cur, "result_output", -1);

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