diff --git a/gpttype_adapter.cpp b/gpttype_adapter.cpp
index 64cfc9055..87d536ed9 100644
--- a/gpttype_adapter.cpp
+++ b/gpttype_adapter.cpp
@@ -2258,6 +2258,8 @@ ModelLoadResult gpttype_load_model(const load_model_inputs inputs, FileFormat in
         std::vector<llama_model_kv_override> kvos; //ensure it keeps in scope until model is created
         std::vector<llama_model_tensor_buft_override> tenos; //ensure it keeps in scope until model is created
         std::vector<std::string> temp_tensor_names; //store temp tensor names to have mem references.
+        temp_tensor_names.reserve(32); //very important, prevents vector from reallocating
+        tenos.reserve(32);
         if(inputs.moe_experts>0)
         {
             printf("\nOverriding number of experts to %d\n",inputs.moe_experts);