graph : remove redundant scale_w parameter (#20235)

src/models/bailingmoe.cpp

@@ -1,6 +1,5 @@
 #include "models.h"
 
-
 llm_build_bailingmoe::llm_build_bailingmoe(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
     ggml_tensor * cur;
     ggml_tensor * inpL;
@@ -97,7 +96,7 @@ llm_build_bailingmoe::llm_build_bailingmoe(const llama_model & model, const llm_
                     nullptr,
                     n_expert, n_expert_used,
                     LLM_FFN_SILU, hparams.expert_weights_norm,
-                    false, hparams.expert_weights_scale,
+                    hparams.expert_weights_scale,
                     LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
                     il);
         cb(moe_out, "ffn_moe_out", il);