echodict/llama.cpp

1#include "models.h"2 3llm_build_plamo::llm_build_plamo(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        // norm23        cur = build_norm(inpL,24                model.layers[il].attn_norm, NULL,25                LLM_NORM_RMS, il);26        cb(cur, "attn_norm", il);27 28        ggml_tensor * sa_inp = cur;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_embd_head, 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_embd_head, 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, NULL, 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            sa_inp = ggml_get_rows(ctx0, sa_inp, inp_out_ids);59            inpL   = ggml_get_rows(ctx0,   inpL, inp_out_ids);60        }61        ggml_tensor * sa_out = cur;62 63        cur = sa_inp;64 65        // feed-forward network66        {67            cur = build_ffn(cur,68                    model.layers[il].ffn_up,   NULL, NULL,69                    model.layers[il].ffn_gate, NULL, NULL,70                    model.layers[il].ffn_down, NULL, NULL,71                    NULL,72                    LLM_FFN_SILU, LLM_FFN_PAR, il);73            cb(cur, "ffn_out", il);74        }75        cur = ggml_add(ctx0, cur, sa_out);76        cur = ggml_add(ctx0, cur, inpL);77 78        cur = build_cvec(cur, il);79        cb(cur, "l_out", il);80 81        // input for next layer82        inpL = cur;83    }84    cur = inpL;85 86    cur = build_norm(cur,87            model.output_norm, NULL,88            LLM_NORM_RMS, -1);89 90    cb(cur, "result_norm", -1);91    res->t_embd = cur;92 93    // lm_head94    cur = build_lora_mm(model.output, cur);95 96    cb(cur, "result_output", -1);97    res->t_logits = cur;98 99    ggml_build_forward_expand(gf, cur);100}101