Brunobkr/llama.cpp_AlgMor24_github

1#include "models.h"2 3void llama_model_minimax_m2::load_arch_hparams(llama_model_loader & ml) {4    ml.get_key(LLM_KV_ATTENTION_LAYERNORM_RMS_EPS,  hparams.f_norm_rms_eps);5    ml.get_key(LLM_KV_EXPERT_FEED_FORWARD_LENGTH,   hparams.n_ff_exp);6    ml.get_key(LLM_KV_EXPERT_GATING_FUNC,           hparams.expert_gating_func, false);7 8    switch (hparams.n_layer()) {9        case 62: type = LLM_TYPE_230B_A10B; break;10        default: type = LLM_TYPE_UNKNOWN;11    }12}13 14void llama_model_minimax_m2::load_arch_tensors(llama_model_loader &) {15    LLAMA_LOAD_LOCALS;16 17    tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);18 19    // output20    output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);21    output      = create_tensor(tn(LLM_TENSOR_OUTPUT,      "weight"), {n_embd, n_vocab}, 0);22 23    for (int i = 0; i < n_layer; ++i) {24        auto & layer = layers[i];25 26        create_tensor_qkv(layer, i, n_embd, n_embd_head_k * n_head, n_embd_gqa, n_embd_gqa, 0);27        layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), { n_embd_head_k * n_head, n_embd }, 0);28 29        layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);30        layer.attn_q_norm = create_tensor(tn(LLM_TENSOR_ATTN_Q_NORM, "weight", i), {n_embd_head_k * n_head}, 0);31        layer.attn_k_norm = create_tensor(tn(LLM_TENSOR_ATTN_K_NORM, "weight", i), {n_embd_k_gqa}, 0);32 33        layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);34 35        layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP, "weight", i), {n_embd, n_expert}, 0);36        layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), {n_embd, n_ff,   n_expert}, 0);37        layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), {n_ff,   n_embd, n_expert}, 0);38        layer.ffn_up_exps   = create_tensor(tn(LLM_TENSOR_FFN_UP_EXPS,   "weight", i), {n_embd, n_ff,   n_expert}, 0);39        layer.ffn_exp_probs_b = create_tensor(tn(LLM_TENSOR_FFN_EXP_PROBS_B, "bias", i), {n_expert}, 0);40    }41}42 43std::unique_ptr<llm_graph_context> llama_model_minimax_m2::build_arch_graph(const llm_graph_params & params) const {44    return std::make_unique<graph>(*this, params);45}46 47llama_model_minimax_m2::graph::graph(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {48    const int64_t n_embd_head = hparams.n_embd_head_v();49 50    GGML_ASSERT(n_embd_head == hparams.n_embd_head_k());51    // GGML_ASSERT(n_embd_head == n_rot); this is wrong in case of minimax, head_dim = 128, n_rot = 6452 53    ggml_tensor * cur;54    ggml_tensor * inpL;55 56    inpL = build_inp_embd(model.tok_embd);57 58    ggml_tensor * inp_pos = build_inp_pos();59    auto inp_attn = build_attn_inp_kv();60    ggml_tensor * inp_out_ids = build_inp_out_ids();61 62    for (int il = 0; il < n_layer; ++il) {63        res->t_layer_inp[il] = inpL;64 65        ggml_tensor * inpSA = inpL;66 67        cur = inpL;68 69        // self_attention70        {71            cur = build_norm(inpL, model.layers[il].attn_norm, NULL, LLM_NORM_RMS, il);72            cb(cur, "attn_norm", il);73 74            // compute Q and K and RoPE them75            ggml_tensor * Qcur = build_lora_mm(model.layers[il].wq, cur);76            cb(Qcur, "Qcur", il);77 78            ggml_tensor * Kcur = build_lora_mm(model.layers[il].wk, cur);79            cb(Kcur, "Kcur", il);80 81            ggml_tensor * Vcur = build_lora_mm(model.layers[il].wv, cur);82            cb(Vcur, "Vcur", il);83 84            Qcur = build_norm(Qcur, model.layers[il].attn_q_norm, NULL,85                    LLM_NORM_RMS, il);86            cb(Qcur, "Qcur_normed", il);87 88            Kcur = build_norm(Kcur, model.layers[il].attn_k_norm, NULL,89                    LLM_NORM_RMS, il);90            cb(Kcur, "Kcur_normed", il);91 92            Qcur = ggml_reshape_3d(ctx0, Qcur, n_embd_head, n_head,    n_tokens);93            Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);94            Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);95 96            Qcur = ggml_rope_ext(97                ctx0, Qcur, inp_pos, nullptr,98                n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,99                ext_factor, attn_factor, beta_fast, beta_slow100                );101 102            Kcur = ggml_rope_ext(103                ctx0, Kcur, inp_pos, nullptr,104                n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,105                ext_factor, attn_factor, beta_fast, beta_slow106                );107 108            cb(Qcur, "Qcur", il);109            cb(Kcur, "Kcur", il);110            cb(Vcur, "Vcur", il);111 112            cur = build_attn(inp_attn,113                    model.layers[il].wo, NULL, model.layers[il].wo_s,114                    Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);115        }116 117        if (il == n_layer - 1 && inp_out_ids) {118            cur   = ggml_get_rows(ctx0,   cur, inp_out_ids);119            inpSA = ggml_get_rows(ctx0, inpSA, inp_out_ids);120        }121 122        ggml_tensor * ffn_inp = ggml_add(ctx0, cur, inpSA);123        cb(ffn_inp, "ffn_inp", il);124 125        // MoE branch126        cur = build_norm(ffn_inp,127                model.layers[il].ffn_norm, NULL,128                LLM_NORM_RMS, il);129        cb(cur, "ffn_norm", il);130 131        cur = build_moe_ffn(cur,132                model.layers[il].ffn_gate_inp,133                model.layers[il].ffn_up_exps,134                model.layers[il].ffn_gate_exps,135                model.layers[il].ffn_down_exps,136                model.layers[il].ffn_exp_probs_b,137                n_expert, n_expert_used,138                LLM_FFN_SILU, true,139                hparams.expert_weights_scale,140                (llama_expert_gating_func_type) hparams.expert_gating_func,141                il);142        cb(cur, "ffn_moe_out", il);143 144        cur = ggml_add(ctx0, cur, ffn_inp);145 146        cur = build_cvec(cur, il);147        cb(cur, "l_out", il);148 149        // input for next layer150        inpL = cur;151    }152 153    cur = inpL;154 155    cur = build_norm(cur,156            model.output_norm, NULL,157            LLM_NORM_RMS, -1);158 159    cb(cur, "result_norm", -1);160    res->t_embd = cur;161 162    // lm_head163    cur = build_lora_mm(model.output, cur, model.output_s);164 165    cb(cur, "result_output", -1);166    res->t_logits = cur;167 168    ggml_build_forward_expand(gf, cur);169}170