Maybe use the proper gating function? Nooo....

Author: pwilkin

src/llama-model.cpp

@@ -19775,7 +19775,7 @@ struct llm_build_minimax_m2 : public llm_graph_context {
                     n_expert, n_expert_used,
                     LLM_FFN_SILU, true,
                     false, 0.0,
-                    LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
+                    (llama_expert_gating_func_type) hparams.expert_gating_func,
                     il);
             cb(cur, "ffn_moe_out", il);