support ppo score norm (trl 0.5.1.dev required)

Former-commit-id: 2b25db6d260ec1532281a592e873579346c7d21c

src/llmtuner/hparams/finetuning_args.py

@@ -61,6 +61,10 @@ class FinetuningArguments:
         default=True,
         metadata={"help": "Whether to resume training from the last LoRA weights or create new weights after merging them."}
     )
+    ppo_score_norm: Optional[bool] = field(
+        default=False,
+        metadata={"help": "Use score normalization in PPO Training."}
+    )
     dpo_beta: Optional[float] = field(
         default=0.1,
         metadata={"help": "The beta parameter for the DPO loss."}