echodict/llama.cpp

1#include "models.h"2 3llm_build_gptneox::llm_build_gptneox(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 8    ggml_tensor * cur;9    ggml_tensor * inpL;10 11    inpL = build_inp_embd(model.tok_embd);12 13    // inp_pos - contains the positions14    ggml_tensor * inp_pos = build_inp_pos();15 16    auto * inp_attn = build_attn_inp_kv();17 18    ggml_tensor * inp_out_ids = build_inp_out_ids();19 20    for (int il = 0; il < n_layer; ++il) {21        cur = build_norm(inpL,22                model.layers[il].attn_norm,23                model.layers[il].attn_norm_b,24                LLM_NORM, il);25        cb(cur, "attn_norm", il);26 27        // self-attention28        {29            auto [Qcur, Kcur, Vcur] = build_qkv(model.layers[il], cur,30                    n_embd_head, n_head, n_head_kv, il);31 32            Qcur = ggml_rope_ext(33                    ctx0, Qcur, inp_pos, nullptr,34                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,35                    ext_factor, attn_factor, beta_fast, beta_slow36                    );37 38            Kcur = ggml_rope_ext(39                    ctx0, Kcur, inp_pos, nullptr,40                    n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,41                    ext_factor, attn_factor, beta_fast, beta_slow42                    );43 44            cb(Qcur, "Qcur", il);45            cb(Kcur, "Kcur", il);46            cb(Vcur, "Vcur", il);47 48            cur = build_attn(inp_attn,49                    model.layers[il].wo, model.layers[il].wo_b, model.layers[il].wo_s,50                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);51        }52 53        if (il == n_layer - 1 && inp_out_ids) {54            cur  = ggml_get_rows(ctx0,  cur, inp_out_ids);55            inpL = ggml_get_rows(ctx0, inpL, inp_out_ids);56        }57 58        // ffn59        if (hparams.use_par_res) {60            // attention and ffn are computed in parallel61            // x = x + attn(ln1(x)) + ffn(ln2(x))62 63            ggml_tensor * attn_out = cur;64 65            cur = build_norm(inpL,66                    model.layers[il].ffn_norm,67                    model.layers[il].ffn_norm_b,68                    LLM_NORM, il);69            cb(cur, "ffn_norm", il);70 71            cur = build_ffn(cur,72                    model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,73                    NULL,                      NULL,                        NULL,74                    model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,75                    NULL,76                    LLM_FFN_GELU, LLM_FFN_SEQ, il);77            cb(cur, "ffn_out", il);78 79            cur = ggml_add(ctx0, cur, inpL);80            cb(cur, "ffn_out", il);81 82            cur = ggml_add(ctx0, cur, attn_out);83 84            cur = build_cvec(cur, il);85            cb(cur, "l_out", il);86 87            // input for next layer88            inpL = cur;89        } else {90            // attention and ffn are computed sequentially91            // x = x + attn(ln1(x))92            // x = x + ffn(ln2(x))93 94            ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpL);95            cb(ffn_inp, "ffn_inp", il);96 97            cur = build_norm(ffn_inp,98                    model.layers[il].ffn_norm,99                    model.layers[il].ffn_norm_b,100                    LLM_NORM, il);101            cb(cur, "ffn_norm", il);102 103            cur = build_ffn(cur,104                    model.layers[il].ffn_up,   model.layers[il].ffn_up_b,   NULL,105                    NULL,                      NULL,                        NULL,106                    model.layers[il].ffn_down, model.layers[il].ffn_down_b, NULL,107                    NULL,108                    LLM_FFN_GELU, LLM_FFN_SEQ, il);109            cb(cur, "ffn_out", il);110 111            cur = ggml_add(ctx0, cur, ffn_inp);112 113            cur = build_cvec(cur, il);114            cb(cur, "l_out", il);115 116            // input for next layer117            inpL = cur;118        }119    }120 121    cur = build_norm(inpL,122            model.output_norm,123            model.output_norm_b,124            LLM_NORM, -1);125 126    cb(cur, "result_norm", -1);127    res->t_embd = cur;128 129    cur = build_lora_mm(model.output, cur);130 131    cb(cur, "result_output", -1);132    res->t_logits = cur;133 134    ggml_build_forward_expand(gf, cur);135}136