add rm center_rewards_coefficient argument (#3917)

hjh0119 · web-flow · commit 19d524ad845f · 2025-04-17T18:44:36.000+08:00
diff --git a/swift/llm/argument/rlhf_args.py b/swift/llm/argument/rlhf_args.py
@@ -98,6 +98,8 @@ class RLHFArguments(GRPOArguments, PPOArguments, RewardModelArguments, TrainArgu
     undesirable_weight: float = 1.0
     # PPO/GRPO
     temperature: float = 0.9
+    # RM
+    center_rewards_coefficient: Optional[float] = None
 
     def _prepare_training_args(self, training_args: Dict[str, Any]) -> None:
         if self.rlhf_type == 'ppo':
