[RL] add lm_head_fp32 in RolloutModelConfig (#5825)

fastdeploy/rl/rollout_config.py

@@ -67,6 +67,7 @@ class RolloutModelConfig:
         eplb_config: str = {},
         routing_replay_config: str = None,
         load_choices: str = "default_v1",
+        lm_head_fp32: bool = False,
     ):
         # Required parameters
         self.model = model_name_or_path
@@ -117,6 +118,7 @@ class RolloutModelConfig:
         self.eplb_config = eplb_config
         self.routing_replay_config = routing_replay_config
         self.load_choices = load_choices
+        self.lm_head_fp32 = lm_head_fp32
 
     def __str__(self):
         return "\n".join(f"{k}: {v}" for k, v in self.__dict__.items())