echodict/llama.cpp

1#include "models.h"2 3llm_build_qwen2moe::llm_build_qwen2moe(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {4    const int64_t n_embd_head = hparams.n_embd_head_v();5 6    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k());7    GGML_ASSERT(n_embd_head == n_rot);8 9    ggml_tensor * cur;10    ggml_tensor * inpL;11 12    inpL = build_inp_embd(model.tok_embd);13 14    // inp_pos - contains the positions15    ggml_tensor * inp_pos = build_inp_pos();16 17    auto * inp_attn = build_attn_inp_kv();18 19    ggml_tensor * inp_out_ids = build_inp_out_ids();20 21    for (int il = 0; il < n_layer; ++il) {22        ggml_tensor * inpSA = inpL;23 24        // norm25        cur = build_norm(inpL,26                model.layers[il].attn_norm, NULL,27                LLM_NORM_RMS, il);28        cb(cur, "attn_norm", il);29 30        // self_attention31        {32            // compute Q and K and RoPE them33            auto [Qcur, Kcur, Vcur] = build_qkv(model.layers[il], cur,34                    n_embd_head, n_head, n_head_kv, il);35 36            Qcur = ggml_rope_ext(37                    ctx0, Qcur, inp_pos, nullptr,38                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,39                    ext_factor, attn_factor, beta_fast, beta_slow40                    );41 42            Kcur = ggml_rope_ext(43                    ctx0, Kcur, inp_pos, nullptr,44                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,45                    ext_factor, attn_factor, beta_fast, beta_slow46                    );47 48            cb(Qcur, "Qcur", il);49            cb(Kcur, "Kcur", il);50            cb(Vcur, "Vcur", il);51 52            cur = build_attn(inp_attn,53                    model.layers[il].wo, model.layers[il].wo_b, model.layers[il].wo_s,54                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);55        }56        if (il == n_layer - 1 && inp_out_ids) {57            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);58            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);59        }60        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);61        cb(ffn_inp, "ffn_inp", il);62 63        // MoE branch64        cur = build_norm(ffn_inp,65                model.layers[il].ffn_norm, NULL,66                LLM_NORM_RMS, il);67        cb(cur, "ffn_norm", il);68 69        ggml_tensor * moe_out =70            build_moe_ffn(cur,71                    model.layers[il].ffn_gate_inp,72                    model.layers[il].ffn_up_exps,73                    model.layers[il].ffn_gate_exps,74                    model.layers[il].ffn_down_exps,75                    nullptr,76                    n_expert, n_expert_used,77                    LLM_FFN_SILU, false,78                    hparams.expert_weights_scale,79                    LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,80                    il);81        cb(moe_out, "ffn_moe_out", il);82 83        // FFN shared expert84        {85            ggml_tensor * cur_gate_inp = build_lora_mm(model.layers[il].ffn_gate_inp_shexp, cur);86            cb(cur_gate_inp, "ffn_shexp_gate_inp", il);87 88            // sigmoid89            ggml_tensor * cur_gate = ggml_div(ctx0, ggml_silu(ctx0, cur_gate_inp), cur_gate_inp);90            cb(cur_gate, "ffn_shexp_gate", il);91 92            ggml_tensor * cur_ffn = build_ffn(cur,93                    model.layers[il].ffn_up_shexp,   NULL, NULL,94                    model.layers[il].ffn_gate_shexp, NULL, NULL,95                    model.layers[il].ffn_down_shexp, NULL, NULL,96                    NULL,97                    LLM_FFN_SILU, LLM_FFN_PAR, il);98            cb(cur_ffn, "ffn_shexp", il);99 100            ggml_tensor * ffn_shexp_out = ggml_mul(ctx0, cur_ffn, cur_gate);101            cb(ffn_shexp_out, "ffn_shexp_out", il);102 103            moe_out = ggml_add(ctx0, moe_out, ffn_shexp_out);104            cb(moe_out, "ffn_out", il);105 106            cur = moe_out;107        }108        cur = ggml_add(ctx0, cur, ffn_inp);109 110        cur = build_cvec(cur, il);111        cb(cur, "l_out", il);112 113        // input for next layer114        inpL = cur;115    }116    cur = inpL;117 118    cur = build_norm(cur,119            model.output_norm, NULL,120            LLM_NORM_RMS, -1);121 122    cb(cur, "result_norm", -1);123    res->t_embd = cur;124 125    // lm_head126    cur = build_lora_mm(model.output, cur);127 128    cb(cur, "result_output", -1);129    res->t_logits = cur;130 131    ggml_build_forward_expand(gf, cur);132}133