echodict/llama.cpp

1#include "models.h"2 3llm_build_smollm3::llm_build_smollm3(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    const float kq_scale = hparams.f_attention_scale == 0.0f ? 1.0f/sqrtf(float(n_embd_head)) : hparams.f_attention_scale;20 21    ggml_tensor * inp_out_ids = build_inp_out_ids();22 23    for (int il = 0; il < n_layer; ++il) {24        ggml_tensor * inpSA = inpL;25 26        const bool use_rope = (il + 1) % hparams.n_no_rope_layer_step != 0;27 28        // norm29        cur = build_norm(inpL,30                model.layers[il].attn_norm, NULL,31                LLM_NORM_RMS, il);32        cb(cur, "attn_norm", il);33 34        // self-attention35        {36            // compute Q and K and RoPE them37            auto [Qcur, Kcur, Vcur] = build_qkv(model.layers[il], cur,38                    n_embd_head, n_head, n_head_kv, il);39 40            if (use_rope) {41                Qcur = ggml_rope_ext(42                        ctx0, Qcur, 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                Kcur = ggml_rope_ext(48                        ctx0, Kcur, inp_pos, nullptr,49                        n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,50                        ext_factor, attn_factor, beta_fast, beta_slow51                        );52            }53            cb(Qcur, "Qcur", il);54            cb(Kcur, "Kcur", il);55            cb(Vcur, "Vcur", il);56 57            cur = build_attn(inp_attn,58                    model.layers[il].wo, model.layers[il].wo_b, model.layers[il].wo_s,59                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, kq_scale, il);60            cb(cur, "attn_out", il);61        }62        if (il == n_layer - 1 && inp_out_ids) {63            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);64            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);65        }66        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);67        cb(ffn_inp, "ffn_inp", il);68 69        // feed-forward network70        {71            cur = build_norm(ffn_inp,72                    model.layers[il].ffn_norm, NULL,73                    LLM_NORM_RMS, il);74            cb(cur, "ffn_norm", il);75 76            cur = build_ffn(cur,77                    model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,78                    model.layers[il].ffn_gate, model.layers[il].ffn_gate_b, NULL,79                    model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,80                    NULL,81                    LLM_FFN_SILU, LLM_FFN_PAR, il);82            cb(cur, "ffn_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    cur = inpL;94 95    cur = build_norm(cur,96            model.output_norm, NULL,97            LLM_NORM_RMS, -1);98 99    cb(cur, "result_norm", -1);100    res->t_embd = cur;101 102    // lm_head103    cur = build_lora_mm(model.output, cur);104 105    cb(cur, "result_output", -1);106    res->t_logits = cur;107 108    ggml_build_forward_expand(gf, cur);109}110