diff --git a/common/train.cpp b/common/train.cpp
index 99c319253..287fe5d93 100644
--- a/common/train.cpp
+++ b/common/train.cpp
@@ -1115,7 +1115,11 @@ void print_common_train_usage(int /*argc*/, char ** argv, const struct train_par
 
 bool consume_common_train_arg(int argc, char ** argv, int * idx, struct train_params_common * params, bool * invalid_param) {
     int& i = *idx;
-    char * arg = argv[i];
+    std::string arg = argv[i];
+    const std::string arg_prefix = "--";
+    if (arg.compare(0, arg_prefix.size(), arg_prefix) == 0) {
+        std::replace(arg.begin(), arg.end(), '_', '-');
+    }
     if (arg == "--train-data") {
         if (++i >= argc) {
             *invalid_param = true;
diff --git a/examples/finetune/finetune.cpp b/examples/finetune/finetune.cpp
index 308e3d592..17b559a10 100644
--- a/examples/finetune/finetune.cpp
+++ b/examples/finetune/finetune.cpp
@@ -514,8 +514,6 @@ static void init_lora(const struct my_llama_model * model, struct my_llama_lora
     ggml_allocr_free(alloc);
 }
 
-
-
 static void randomize_lora(struct my_llama_lora * lora, int seed, float mean, float std, float min, float max) {
     const uint32_t n_layer = lora->layers.size();