[plugin] Custom model_runner/model support (#3186)

* support custom model&&model_runner

* fix merge

* add test && update doc

* fix codestyle

* fix unittest

* load model in rl

fastdeploy/model_executor/layers/attention/attention.py

@@ -48,6 +48,7 @@ class Attention(nn.Layer):
         linear_shift: paddle.Tensor = None,
         linear_smooth: paddle.Tensor = None,
         use_neox_rotary_style: bool = False,
+        use_qk_norm: bool = False,
     ) -> None:
         """
         Initializes `LMLayer` with the given parameters.

fastdeploy/model_executor/models/model_base.py

@@ -64,9 +64,9 @@ class ModelRegistry:
 
     @classmethod
     def get_supported_archs(cls):
-        assert len(cls._arch_to_model_cls) == len(
-            cls._arch_to_model_cls
-        ), "model class / pretrained model registry num is not same"
+        assert len(cls._arch_to_model_cls) >= len(
+            cls._arch_to_pretrained_model_cls
+        ), "model class num is more than pretrained model registry num"
         return [key for key in cls._arch_to_model_cls.keys()]