[Model]support qwen2_5_vl (#3557)

* adapt qwen_2_5_vl model

* adapt qwen_2_5_vl VIT model

* adapt qwen2_5_vl images_embeds

* adapt qwen2_5_vl 3D rope

* adapt qwen2_5_vl 3D rope v2

* adapt qwen2_5_vl processor

* adapt qwen2_5_vl bypass resampler_model

* adapt qwen2_5_vl 绕过部分ernie逻辑

* adapt qwen2_5_vl 绕过部分ernie逻辑 v2

* adapt qwen2_5_vl 权重加载与命名修改

* adapt qwen2_5_vl 非必须think_end_id

* adapt qwen2_5_vl 区分多种模型的extract_vision_features

* fix:adapt qwen2_5_vl model

* adapt qwen2_5_vl norm

* adapt qwen2_5_vl  processor 更新

* adapt qwen2_5_vl image and video success

* adapt qwen2_5_vl 部分整理代码

* adapt qwen2_5_vl 支持多卡

* adapt qwen2_5_vl on latest develop

* adapt qwen2_5_vl RL

* adapt qwen2_5_vl 整理代码

* support noex rope3d

* adapt qwen2_5_vl add init.py

* adapt qwen2_5_vl add init.py v2

* adapt qwen2_5_vl remove space

* adapt qwen2_5_vl remove space v2

* adapt qwen2_5_vl pre-commit

* adapt qwen2_5_vl update

* adapt qwen2_5_vl pre-commit v2

* adapt qwen2_5_vl modify comments

* adapt qwen2_5_vl fix indentation

* adapt qwen2_5_vl fix indentation v2

---------

Co-authored-by: wangyafeng <wangyafeng@baidu.com>
Co-authored-by: xiaoxiaohehe001 <49090790+xiaoxiaohehe001@users.noreply.github.com>
Co-authored-by: CSWYF3634076 <58356743+CSWYF3634076@users.noreply.github.com>

fastdeploy/rl/rollout_model.py

@@ -33,6 +33,10 @@ from fastdeploy.model_executor.models.qwen2 import (
     Qwen2ForCausalLM,
     Qwen2PretrainedModel,
 )
+from fastdeploy.model_executor.models.qwen2_5_vl.qwen2_5_vl import (
+    Qwen2_5_VLForConditionalGeneration,
+    Qwen2_5_VLPretrainedModel,
+)
 from fastdeploy.model_executor.models.qwen3 import (
     Qwen3ForCausalLM,
     Qwen3PretrainedModel,
@@ -477,3 +481,51 @@ class Qwen3ForCausalLMRL(Qwen3ForCausalLM, BaseRLModel):
         self._complete_missing_mappings()
 
         return self.infer_to_train_mapping
+
+
+class Qwen2_5_VLForConditionalGenerationRL(Qwen2_5_VLForConditionalGeneration, BaseRLModel):
+    """
+    Qwen2_5_VLForConditionalGenerationRL
+    """
+
+    _get_tensor_parallel_mappings = Qwen2_5_VLPretrainedModel._get_tensor_parallel_mappings
+
+    def __init__(self, fd_config: FDConfig):
+        """
+        Args:
+            fd_config (FDConfig): Configurations for the LLM model.
+        """
+        super(Qwen2_5_VLForConditionalGenerationRL, self).__init__(fd_config)
+
+    @classmethod
+    def name(self) -> str:
+        """name"""
+        return "Qwen2_5_VLForConditionalGenerationRL"
+
+    def get_name_mappings_to_training(self, trainer_degree=None) -> Dict[str, str]:
+        if self._mappings_built:
+            return self.infer_to_train_mapping
+
+        self.infer_to_train_mapping = {}
+        self._mappings_built = True
+        # Prepare placeholders
+        place_holders = ["weight"]
+
+        # Initialize mapping dictionary
+        self._update_base_mappings("model")
+        base_name = "model.layers"
+
+        # Helper function to add layer mappings
+        def _add_layer_mappings(layer_idx):
+            # FFN mappings
+            for ph in place_holders:
+                self.infer_to_train_mapping[f"{base_name}.{layer_idx}.mlp.up_gate_proj.{ph}"] = (
+                    f"{base_name}.{layer_idx}.mlp.gate_up_fused_proj.{ph}"
+                )
+
+        for layer_idx in range(self.fd_config.model_config.num_hidden_layers):
+            _add_layer_mappings(layer_idx)
+
+        self._complete_missing_mappings()
+
+        return self.infer_to_train_mapping