| 1056 | } |
| 1057 | |
| 1058 | ggml_tensor * llm_graph_context::build_moe_ffn( |
| 1059 | ggml_tensor * cur, |
| 1060 | ggml_tensor * gate_inp, |
| 1061 | ggml_tensor * up_exps, |
| 1062 | ggml_tensor * gate_exps, |
| 1063 | ggml_tensor * down_exps, |
| 1064 | ggml_tensor * exp_probs_b, |
| 1065 | int64_t n_expert, |
| 1066 | int64_t n_expert_used, |
| 1067 | llm_ffn_op_type type_op, |
| 1068 | bool norm_w, |
| 1069 | bool scale_w, |
| 1070 | float w_scale, |
| 1071 | llama_expert_gating_func_type gating_op, |
| 1072 | int il, |
| 1073 | ggml_tensor * probs_in) const { |
| 1074 | return build_moe_ffn( |
| 1075 | cur, |
| 1076 | gate_inp, /* gate_inp_b */ nullptr, |
| 1077 | up_exps, /* up_exps_b */ nullptr, |
| 1078 | gate_exps, /* gate_exps_b */ nullptr, |
| 1079 | down_exps, /* down_exps_b */ nullptr, |
| 1080 | exp_probs_b, |
| 1081 | n_expert, |
| 1082 | n_expert_used, |
| 1083 | type_op, |
| 1084 | norm_w, |
| 1085 | scale_w, |
| 1086 | w_scale, |
| 1087 | gating_op, |
| 1088 | il, |
| 1089 | probs_in |
| 1090 | ); |
| 1091 | } |
| 1092 | |
| 1093 | ggml_tensor * llm_graph_context::build_moe_ffn( |
| 1094 | ggml_tensor * cur, |
nothing calls this directly
no test coverage detected