| 767 | // |
| 768 | |
| 769 | llm_graph_context::llm_graph_context(const llm_graph_params & params) : |
| 770 | arch (params.arch), |
| 771 | hparams (params.hparams), |
| 772 | cparams (params.cparams), |
| 773 | ubatch (params.ubatch), |
| 774 | n_embd (hparams.n_embd), |
| 775 | n_layer (hparams.n_layer), |
| 776 | n_rot (hparams.n_rot), |
| 777 | n_ctx (cparams.n_ctx), |
| 778 | n_head (hparams.n_head()), |
| 779 | n_head_kv (hparams.n_head_kv()), |
| 780 | n_embd_head_k (hparams.n_embd_head_k), |
| 781 | n_embd_k_gqa (hparams.n_embd_k_gqa()), |
| 782 | n_embd_head_v (hparams.n_embd_head_v), |
| 783 | n_embd_v_gqa (hparams.n_embd_v_gqa()), |
| 784 | n_expert (hparams.n_expert), |
| 785 | n_expert_used (cparams.warmup ? hparams.n_expert : hparams.n_expert_used), |
| 786 | freq_base (cparams.rope_freq_base), |
| 787 | freq_scale (cparams.rope_freq_scale), |
| 788 | ext_factor (cparams.yarn_ext_factor), |
| 789 | attn_factor (cparams.yarn_attn_factor), |
| 790 | beta_fast (cparams.yarn_beta_fast), |
| 791 | beta_slow (cparams.yarn_beta_slow), |
| 792 | norm_eps (hparams.f_norm_eps), |
| 793 | norm_rms_eps (hparams.f_norm_rms_eps), |
| 794 | n_tokens (ubatch.n_tokens), |
| 795 | n_outputs (params.n_outputs), |
| 796 | n_ctx_orig (cparams.n_ctx_orig_yarn), |
| 797 | pooling_type (cparams.pooling_type), |
| 798 | rope_type (hparams.rope_type), |
| 799 | sched (params.sched), |
| 800 | backend_cpu (params.backend_cpu), |
| 801 | cvec (params.cvec), |
| 802 | loras (params.loras), |
| 803 | mctx (params.mctx), |
| 804 | cross (params.cross), |
| 805 | samplers (params.samplers), |
| 806 | cb_func (params.cb), |
| 807 | res (params.res), |
| 808 | ctx0 (res->get_ctx()), |
| 809 | gf (res->get_gf()) { |
| 810 | res->set_params(params); |
| 811 | } |
| 812 | |
| 813 | void llm_graph_context::cb(ggml_tensor * cur, const char * name, int il) const { |
| 814 | if (cb_func) { |
nothing calls this directly
no test coverage detected