diff --git a/gpttype_adapter.cpp b/gpttype_adapter.cpp
index 02c7d56b0..edb7f0206 100644
--- a/gpttype_adapter.cpp
+++ b/gpttype_adapter.cpp
@@ -970,9 +970,9 @@ ModelLoadResult gpttype_load_model(const load_model_inputs inputs, FileFormat in
         model_params.main_gpu = cu_parseinfo_maindevice;
 
         #if defined(GGML_USE_CUBLAS)
-        model_params.split_mode = (inputs.use_rowsplit?llama_split_mode::LLAMA_SPLIT_ROW:llama_split_mode::LLAMA_SPLIT_LAYER);
+        model_params.split_mode = (inputs.use_rowsplit?llama_split_mode::LLAMA_SPLIT_MODE_ROW:llama_split_mode::LLAMA_SPLIT_MODE_LAYER);
         #else
-        model_params.split_mode = llama_split_mode::LLAMA_SPLIT_LAYER;
+        model_params.split_mode = llama_split_mode::LLAMA_SPLIT_MODE_LAYER;
         #endif
 
         llama_ctx_params.n_batch = kcpp_params->n_batch;
diff --git a/llama.cpp b/llama.cpp
index ffc45f4bc..cac6b6271 100644
--- a/llama.cpp
+++ b/llama.cpp
@@ -2863,7 +2863,7 @@ struct llama_model_loader {
                 bool shouldoffload = (layernum>=0 && clblast_offload_fallback_layers>layernum);
                 if(shouldoffload)
                 {
-                    cur->backend = GGML_BACKEND_GPU;
+                    cur->backend = GGML_BACKEND_TYPE_GPU;
                     ggml_cl_transform_tensor(cur->data, cur);
                 }
             }