diff --git a/swift/plugin/orm.py b/swift/plugin/orm.py
index cf8745de74..b43cd401b2 100644
--- a/swift/plugin/orm.py
+++ b/swift/plugin/orm.py
@@ -296,12 +296,14 @@ def __call__(self, completions, **kwargs) -> List[float]:
 class CosineReward(ORM):
     # https://arxiv.org/abs/2502.03373
     def __init__(self,
+                 tokenizer=None,
                  cosine_min_len_value_wrong: float = 0.0,
                  cosine_max_len_value_wrong: float = -0.5,
                  cosine_min_len_value_correct: float = 1.0,
                  cosine_max_len_value_correct: float = 0.5,
                  cosine_max_len: int = 1000,
                  accuracy_orm=None):
+        self.tokenizer = tokenizer
         self.min_len_value_wrong = cosine_min_len_value_wrong
         self.max_len_value_wrong = cosine_max_len_value_wrong
         self.min_len_value_correct = cosine_min_len_value_correct
@@ -326,7 +328,7 @@ def __call__(self, completions, solution, **kwargs) -> List[float]:
             else:
                 min_value = self.min_len_value_wrong
                 max_value = self.max_len_value_wrong
-            gen_len = len(content)
+            gen_len = len(self.tokenizer.encode(content))
             reward = self.cosfn(gen_len, self.max_len, min_value, max_value)
             rewards.append(reward)
         return rewards
diff --git a/swift/trainers/rlhf_trainer/grpo_trainer.py b/swift/trainers/rlhf_trainer/grpo_trainer.py
index 1fdf5034d6..25e18e441e 100644
--- a/swift/trainers/rlhf_trainer/grpo_trainer.py
+++ b/swift/trainers/rlhf_trainer/grpo_trainer.py
@@ -71,6 +71,7 @@ def __init__(self,
         self.args = args
         self.queue = Queue()
         self.processing_class = kwargs.get('template').tokenizer
+        self.tokenizer = kwargs.get('template').tokenizer if hasattr(kwargs.get('template'), 'tokenizer') else None
         if not isinstance(reward_funcs, list):
             reward_funcs = [reward_funcs]
 
@@ -83,6 +84,8 @@ def __init__(self,
                         key: getattr(args, key)
                         for key in reward_func_args if key not in ['self', 'args', 'kwargs'] and hasattr(args, key)
                     }
+                    if reward_func_class.__name__ == 'CosineReward' and 'tokenizer' in reward_func_args:
+                        reward_func_kwargs['tokenizer'] = self.tokenizer
                     reward_funcs[i] = reward_func_class(**reward_func_kwargs)
                 elif not callable(reward_func):
                     raise ValueError(f'reward_function {reward_func} is not implemented in swift.llm.plugin')