echodict/llama.cpp

1#include "models.h"2 3llm_build_minimax_m2::llm_build_minimax_m2(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); this is wrong in case of minimax, head_dim = 128, n_rot = 648 9    ggml_tensor * cur;10    ggml_tensor * inpL;11 12    inpL = build_inp_embd(model.tok_embd);13 14    ggml_tensor * inp_pos = build_inp_pos();15    auto inp_attn = build_attn_inp_kv();16    ggml_tensor * inp_out_ids = build_inp_out_ids();17 18    for (int il = 0; il < n_layer; ++il) {19        ggml_tensor * inpSA = inpL;20 21        cur = inpL;22 23        // self_attention24        {25            cur = build_norm(inpL, model.layers[il].attn_norm, NULL, LLM_NORM_RMS, il);26            cb(cur, "attn_norm", il);27 28            // compute Q and K and RoPE them29            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);30            cb(Qcur, "Qcur", il);31 32            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);33            cb(Kcur, "Kcur", il);34 35            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);36            cb(Vcur, "Vcur", il);37 38            Qcur = build_norm(Qcur, model.layers[il].attn_q_norm, NULL,39                    LLM_NORM_RMS, il);40            cb(Qcur, "Qcur_normed", il);41 42            Kcur = build_norm(Kcur, model.layers[il].attn_k_norm, NULL,43                    LLM_NORM_RMS, il);44            cb(Kcur, "Kcur_normed", il);45 46            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);47            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);48            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);49 50            Qcur = ggml_rope_ext(51                ctx0, Qcur, inp_pos, nullptr,52                n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,53                ext_factor, attn_factor, beta_fast, beta_slow54                );55 56            Kcur = ggml_rope_ext(57                ctx0, Kcur, inp_pos, nullptr,58                n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,59                ext_factor, attn_factor, beta_fast, beta_slow60                );61 62            cb(Qcur, "Qcur", il);63            cb(Kcur, "Kcur", il);64            cb(Vcur, "Vcur", il);65 66            cur = build_attn(inp_attn,67                    model.layers[il].wo, NULL, model.layers[il].wo_s,68                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);69        }70 71        if (il == n_layer - 1 && inp_out_ids) {72            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);73            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);74        }75 76        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);77        cb(ffn_inp, "ffn_inp", il);78 79        // MoE branch80        cur = build_norm(ffn_inp,81                model.layers[il].ffn_norm, NULL,82                LLM_NORM_RMS, il);83        cb(cur, "ffn_norm", il);84 85        cur = build_moe_ffn(cur,86                model.layers[il].ffn_gate_inp,87                model.layers[il].ffn_up_exps,88                model.layers[il].ffn_gate_exps,89                model.layers[il].ffn_down_exps,90                model.layers[il].ffn_exp_probs_b,91                n_expert, n_expert_used,92                LLM_FFN_SILU, true,93                hparams.expert_weights_scale,94                (llama_expert_gating_func_type) hparams.expert_gating_func,95                il);96        cb(cur, "ffn_moe_out", il);97 98        cur = ggml_add(ctx0, cur, ffn_inp);99 100        cur = build_cvec(cur, il);101        cb(cur, "l_out", il);102 103        // input for next layer104        inpL = cur;105    }106 107    cur = inpL;108 109    cur = build_norm(cur,110            model.output_norm, NULL,111            LLM_NORM_RMS, -1);112 113    cb(cur, "result_norm", -1);114    res->t_embd = cur;115 116    // lm_head117    cur = build_lora_mm(model.output, cur);118 119    cb(cur, "result_output", -1);120    res->t_logits = cur;121 122    ggml_build_forward_expand(gf, cur);123}124