echodict/llama.cpp

1#include "models.h"2 3llm_build_dbrx::llm_build_dbrx(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, il);28        cb(cur, "attn_norm", il);29 30        // self-attention31        {32            auto [Qcur, Kcur, Vcur] = build_qkv(model.layers[il], cur,33                    n_embd_head, n_head, n_head_kv, il);34 35            Qcur = ggml_rope_ext(36                    ctx0, Qcur, inp_pos, nullptr,37                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,38                    ext_factor, attn_factor, beta_fast, beta_slow39                    );40 41            Kcur = ggml_rope_ext(42                    ctx0, Kcur, inp_pos, nullptr,43                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,44                    ext_factor, attn_factor, beta_fast, beta_slow45                    );46 47            cb(Qcur, "Qcur", il);48            cb(Kcur, "Kcur", il);49            cb(Vcur, "Vcur", il);50 51            cur = build_attn(inp_attn,52                    model.layers[il].wo, NULL, model.layers[il].wo_s,53                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);54        }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 61        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);62        cb(ffn_inp, "ffn_inp", il);63 64        // feed-forward network65        // MoE branch66        cur = build_norm(ffn_inp,67                model.layers[il].attn_out_norm, NULL,68                LLM_NORM, il);69        cb(cur, "attn_out_norm", il);70 71        cur = build_moe_ffn(cur,72                model.layers[il].ffn_gate_inp,73                model.layers[il].ffn_up_exps,74                model.layers[il].ffn_gate_exps,75                model.layers[il].ffn_down_exps,76                nullptr,77                n_expert, n_expert_used,78                LLM_FFN_SILU, true,79                hparams.expert_weights_scale,80                LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,81                il);82        cb(cur, "ffn_moe_out", il);83 84        cur = ggml_add(ctx0, cur, ffn_inp);85        cb(cur, "ffn_out", il);86 87        cur = build_cvec(cur, il);88        cb(cur, "l_out", il);89 90        // input for next layer91        inpL = cur;92    }93 94    cur = inpL;95 96    cur = build_norm(cur,97            model.output_norm, NULL,98            LLM_NORM, -1);99 100    cb(cur, "result_norm", -1);101    res->t_embd = cur;102 103    // lm_head104    cur = build_lora_mm(model.output, cur);105 106    cb(cur, "result_output", -1);107    res->t_logits = cur;108 109    ggml_build_forward_expand(gf, cur);110}111