[MTP]Develop mtp reshard

fastdeploy/config.py

@@ -134,6 +134,7 @@ def __init__(
         self.lm_head_fp32: bool = False
         self.model_format = "auto"
         self.partial_rotary_factor: float = 1.0
+        self.num_nextn_predict_layers = 0
         for key, value in args.items():
             if hasattr(self, key) and value != "None":
                 setattr(self, key, value)

fastdeploy/multimodal/registry.py

@@ -24,6 +24,7 @@ class MultimodalRegistry:
         "Ernie4_5_VLMoeForConditionalGeneration",
         "Ernie5MoeForCausalLM",
         "Qwen2_5_VLForConditionalGeneration",
+        "Ernie5ForCausalLM",
     }
 
     @classmethod

fastdeploy/rl/rollout_config.py

@@ -63,6 +63,7 @@ def __init__(
         local_rank: int = 0,
         moba_attention_config: str = None,
         data_parallel_size: int = 1,
+        num_nextn_predict_layers: int = 0,
     ):
         # Required parameters
         self.model = model_name_or_path
@@ -109,6 +110,7 @@ def __init__(
         self.early_stop_config = early_stop_config
         self.ips = None
         self.moba_attention_config = moba_attention_config
+        self.num_nextn_predict_layers = num_nextn_predict_layers
 
     def __str__(self):
         return "\n".join(f"{k}: {v}" for k, v in self.__dict__.items())