diff --git a/gpttype_adapter.cpp b/gpttype_adapter.cpp
index fdd85b42b..baf861799 100644
--- a/gpttype_adapter.cpp
+++ b/gpttype_adapter.cpp
@@ -3274,7 +3274,7 @@ ModelLoadResult gpttype_load_model(const load_model_inputs inputs, FileFormat in
             // Match llama-server's target rollback slots for speculative verification.
             llama_ctx_params.n_rs_seq = inputs.draft_amount;
         }
-        model_params.load_mode = inputs.use_mlock ? LLAMA_LOAD_MODE_MLOCK : (inputs.use_mmap ? LLAMA_LOAD_MODE_MMAP : LLAMA_LOAD_MODE_NONE);
+        model_params.load_mode = inputs.use_mlock ? (inputs.use_mmap ? LLAMA_LOAD_MODE_MMAP_MLOCK : LLAMA_LOAD_MODE_MLOCK) : (inputs.use_mmap ? LLAMA_LOAD_MODE_MMAP : LLAMA_LOAD_MODE_NONE);
         model_params.n_gpu_layers = inputs.gpulayers;
         model_params.no_host = inputs.no_host;
         model_params.load_mtp = inputs.use_mtp;