diff --git a/trl/experimental/minillm/minillm_config.py b/trl/experimental/minillm/minillm_config.py
index bca7cfb9822..386c44d2948 100644
--- a/trl/experimental/minillm/minillm_config.py
+++ b/trl/experimental/minillm/minillm_config.py
@@ -47,7 +47,7 @@ class MiniLLMConfig(GRPOConfig):
             Whether to apply length normalization to the rewards.
     """
 
-    _VALID_DICT_FIELDS = TrainingArguments._VALID_DICT_FIELDS + ["teacher_model_init_kwargs"]
+    _VALID_DICT_FIELDS = GRPOConfig._VALID_DICT_FIELDS + ["teacher_model_init_kwargs"]
 
     teacher_model_init_kwargs: dict[str, Any] | str | None = field(
         default=None,