@@ -907,15 +907,15 @@ ggml_tensor * llm_graph_context::build_moe_ffn(
907907 cb (cur, " ffn_moe_weighted" , il);
908908 }
909909
910- ggml_tensor * tmp = build_lora_mm_id (up_exps, cur, selected_experts); // [n_ff, n_expert_used, n_tokens]
911- cb (tmp , " ffn_moe_up" , il);
910+ ggml_tensor * up = build_lora_mm_id (up_exps, cur, selected_experts); // [n_ff, n_expert_used, n_tokens]
911+ cb (up , " ffn_moe_up" , il);
912912
913913 ggml_tensor * experts = nullptr ;
914914 if (gate_exps) {
915915 cur = build_lora_mm_id (gate_exps, cur, selected_experts); // [n_ff, n_expert_used, n_tokens]
916916 cb (cur, " ffn_moe_gate" , il);
917917 } else {
918- cur = tmp ;
918+ cur = up ;
919919 }
920920
921921 switch (type_op) {
@@ -934,7 +934,7 @@ ggml_tensor * llm_graph_context::build_moe_ffn(
934934 }
935935
936936 if (gate_exps) {
937- cur = ggml_mul (ctx0, cur, tmp ); // [n_ff, n_expert_used, n_tokens]
937+ cur = ggml_mul (ctx0, cur, up ); // [n_ff, n_expert_used, n_tokens]
938938 cb (cur, " ffn_moe_gate_par" , il);
939939 }
940940
0 commit comments