diff --git a/trl/trainer/grpo_trainer.py b/trl/trainer/grpo_trainer.py
index 267b64b6e3b..9511f27d5e0 100644
--- a/trl/trainer/grpo_trainer.py
+++ b/trl/trainer/grpo_trainer.py
@@ -1259,12 +1259,15 @@ def _generate_single_turn(self, prompts: list):
             processor_kwargs = {
                 "max_length": self.max_prompt_length,
                 "truncation": True,
-                "add_generation_prompt": True,
                 "add_special_tokens": False,
             }
             if is_conversational({"prompt": prompts[0]}):
                 processor_outputs = self.processing_class.apply_chat_template(
-                    conversation=prompts, **processor_kwargs, tokenize=True, return_dict=True
+                    conversation=prompts,
+                    **processor_kwargs,
+                    add_generation_prompt=True,
+                    tokenize=True,
+                    return_dict=True,
                 )
             else:
                 processor_outputs = self.processing_class(text=prompts, **processor_kwargs)
diff --git a/trl/trainer/rloo_trainer.py b/trl/trainer/rloo_trainer.py
index 3ed56624e83..b9fdb3fd5b1 100644
--- a/trl/trainer/rloo_trainer.py
+++ b/trl/trainer/rloo_trainer.py
@@ -1084,13 +1084,13 @@ def _generate_single_turn(self, prompts: list):
             processor_kwargs = {
                 "max_length": self.max_prompt_length,
                 "truncation": True,
-                "add_generation_prompt": True,
                 "add_special_tokens": False,
             }
             if is_conversational({"prompt": prompts[0]}):
                 processor_outputs = self.processing_class.apply_chat_template(
                     conversation=prompts,
                     **processor_kwargs,
+                    add_generation_prompt=True,
                     tokenize=True,
                     return_dict=True,
                 )
@@ -1133,7 +1133,7 @@ def _generate_single_turn(self, prompts: list):
                 generate_inputs = self.processing_class.apply_chat_template(
                     conversation=prompts,
                     **processor_kwargs,
-                    add_generation_kwargs=True,
+                    add_generation_prompt=True,
                     tokenize=True,
                     return_dict=True,
                 )