diff --git a/eval/evaluate_chat_ceval.py b/eval/evaluate_chat_ceval.py
index b909a6d..1c0f70d 100644
--- a/eval/evaluate_chat_ceval.py
+++ b/eval/evaluate_chat_ceval.py
@@ -31,6 +31,7 @@ def load_models_tokenizer(args):
         args.checkpoint_path, trust_remote_code=True
     )
     model.generation_config.do_sample = False  # use greedy decoding
+    model.generation_config.repetition_penalty = 1.0  # disable repetition penalty
     return model, tokenizer
 
 def process_before_extraction(gen, question, choice_dict):
diff --git a/eval/evaluate_chat_gsm8k.py b/eval/evaluate_chat_gsm8k.py
index c4de01e..f4cc22e 100644
--- a/eval/evaluate_chat_gsm8k.py
+++ b/eval/evaluate_chat_gsm8k.py
@@ -129,6 +129,7 @@ if __name__ == "__main__":
         args.checkpoint_path, trust_remote_code=True
     )
     model.generation_config.do_sample = False  # use greedy decoding
+    model.generation_config.repetition_penalty = 1.0  # disable repetition penalty
 
     test = dataset["test"]
 
diff --git a/eval/evaluate_chat_humaneval.py b/eval/evaluate_chat_humaneval.py
index 54ceca8..134a825 100644
--- a/eval/evaluate_chat_humaneval.py
+++ b/eval/evaluate_chat_humaneval.py
@@ -93,6 +93,7 @@ if __name__ == "__main__":
         args.checkpoint_path, trust_remote_code=True
     )
     model.generation_config.do_sample = False  # use greedy decoding
+    model.generation_config.repetition_penalty = 1.0  # disable repetition penalty
 
     f_output = jsonlines.Writer(open(args.sample_output_file, "w", encoding="utf-8"))
 
diff --git a/eval/evaluate_chat_mmlu.py b/eval/evaluate_chat_mmlu.py
index 259dc3a..36d0524 100644
--- a/eval/evaluate_chat_mmlu.py
+++ b/eval/evaluate_chat_mmlu.py
@@ -35,6 +35,7 @@ def load_models_tokenizer(args):
         args.checkpoint_path, trust_remote_code=True
     )
     model.generation_config.do_sample = False  # use greedy decoding
+    model.generation_config.repetition_penalty = 1.0  # disable repetition penalty
     return model, tokenizer