prevent segfault on load fail

expose.cpp

@@ -35,6 +35,11 @@ extern "C"
         draftmodel_filename = inputs.draftmodel_filename;
 
         file_format = check_file_format(model.c_str(),&file_format_meta);
+        if (file_format == FileFormat::BADFORMAT)
+        {
+            fprintf(stderr, "%s: error: invalid or unsupported model file '%s'\n", __func__, model.c_str());
+            return false;
+        }
 
         executable_path = inputs.executable_path;
 

gpttype_adapter.cpp

@@ -3564,6 +3564,11 @@ ModelLoadResult gpttype_load_model(const load_model_inputs inputs, FileFormat in
         }
 
         llama_model * llamamodel = llama_model_load_from_file(kcpp_data->model_filename.c_str(), model_params);
+        if (llamamodel == nullptr)
+        {
+            fprintf(stderr, "%s: error: failed to load model '%s'\n", __func__, kcpp_data->model_filename.c_str());
+            return ModelLoadResult::FAIL;
+        }
 
         //now that the model is loaded, immediately check if SWA is used
         bool model_has_swa = (llama_model_n_swa(llamamodel)!=0);

koboldcpp.py

@@ -12360,6 +12360,10 @@ def kcpp_main_process(launch_args, g_memory=None, gui_launcher=False):
             print("WARNING: Selected Text Model does not seem to be a GGUF file! Are you sure you picked the right file?")
         loadok = load_model(modelname)
         print("Load Text Model OK: " + str(loadok))
+        if not loadok:
+            exitcounter = 999
+            exit_with_error(3,"Could not load text model: " + modelname)
+
         if args.mmproj and args.mmproj!="": # multimodal vision and audio support is only known at runtime
             has_audio_support = handle.has_audio_support()
             has_vision_support = handle.has_vision_support()
@@ -12367,10 +12371,6 @@ def kcpp_main_process(launch_args, g_memory=None, gui_launcher=False):
             has_audio_support = False
             has_vision_support = False
 
-        if not loadok:
-            exitcounter = 999
-            exit_with_error(3,"Could not load text model: " + modelname)
-
         # The chat completions adapter is a list that needs derivation from chat templates
         # Try to derive chat completions adapter from chat template, now that we have the model loaded
         if args.model_param:

model_adapter.cpp

@@ -322,6 +322,11 @@ std::string gguf_get_model_arch(const std::string & gguf_filename)
         ggufparams.ctx = NULL;
 
         auto ctx  = gguf_init_from_file(fname.c_str(), ggufparams);
+        if (ctx == nullptr)
+        {
+            fprintf(stderr, "%s: error: failed to read GGUF file '%s'\n", __func__, fname.c_str());
+            return FileFormat::BADFORMAT;
+        }
 
         auto keyidx = gguf_find_key(ctx, "general.architecture");
         std::string modelarch = "";