diff --git a/mmengine/_strategy/base.py b/mmengine/_strategy/base.py
index 5df3a79c92..05d070f6f2 100644
--- a/mmengine/_strategy/base.py
+++ b/mmengine/_strategy/base.py
@@ -322,7 +322,8 @@ def compile_model(
         Returns:
             nn.Module: Compiled model.
         """
-        if isinstance(compile, bool) and not compile:
+        if  isinstance(compile, bool) and not compile or \
+            isinstance(compile, dict) and not compile.get('disable', False):
             return model
 
         assert digit_version(TORCH_VERSION) >= digit_version('2.0.0'), (
diff --git a/mmengine/_strategy/fsdp.py b/mmengine/_strategy/fsdp.py
index 0788fafdab..124dfd7c57 100644
--- a/mmengine/_strategy/fsdp.py
+++ b/mmengine/_strategy/fsdp.py
@@ -408,7 +408,7 @@ def load_optim_state_dict(self, state_dict: dict) -> None:
                 ``optimizer.state_dict()``
         """
         optim_state_dict = FSDP.optim_state_dict_to_load(
-            state_dict, self.model, self.optim_wrapper.optimizer)
+            self.model, self.optim_wrapper.optimizer, state_dict)
         self.optim_wrapper.load_state_dict(optim_state_dict)
 
     def _init_state_dict_cfg(self, state_dict_cfg: Union[str, dict]) -> None:
diff --git a/mmengine/optim/optimizer/amp_optimizer_wrapper.py b/mmengine/optim/optimizer/amp_optimizer_wrapper.py
index 4f3323f2cc..60200924b5 100644
--- a/mmengine/optim/optimizer/amp_optimizer_wrapper.py
+++ b/mmengine/optim/optimizer/amp_optimizer_wrapper.py
@@ -1,5 +1,6 @@
 # Copyright (c) OpenMMLab. All rights reserved.
 from contextlib import contextmanager
+from functools import partial
 from typing import Union
 
 import torch
@@ -17,7 +18,8 @@
 elif is_mlu_available():
     from torch.mlu.amp import GradScaler
 else:
-    from torch.cuda.amp import GradScaler
+    from torch.amp import GradScaler as amp_GradScaler
+    GradScaler = partial(amp_GradScaler, device='cuda')
 
 
 @OPTIM_WRAPPERS.register_module()
diff --git a/mmengine/runner/runner.py b/mmengine/runner/runner.py
index 68716ab253..7160ac84d7 100644
--- a/mmengine/runner/runner.py
+++ b/mmengine/runner/runner.py
@@ -1,5 +1,6 @@
 # Copyright (c) OpenMMLab. All rights reserved.
 import copy
+import inspect
 import logging
 import os
 import os.path as osp
@@ -902,8 +903,18 @@ def wrap_model(
                 find_unused_parameters=find_unused_parameters)
         else:
             model_wrapper_cfg.setdefault('type', 'MMDistributedDataParallel')
-            model_wrapper_type = MODEL_WRAPPERS.get(
-                model_wrapper_cfg.get('type'))  # type: ignore
+            model_wrapper_type = model_wrapper_cfg.get('type')
+            if isinstance(model_wrapper_type, str):
+                model_wrapper_type = MODEL_WRAPPERS.get(model_wrapper_type)  # type: ignore
+            elif inspect.isclass(model_wrapper_type):
+                pass
+            else:
+                raise KeyError(
+                        f'{model_wrapper_type} is not in the '
+                        'registry. Please check whether the value of '
+                        f'`{model_wrapper_type}` is correct or it was registered '
+                        'as expected. More details can be found at https://mmengine.readthedocs.io/en/latest/advanced_tutorials/config.html#import-the-custom-module'  # noqa: E501
+                    )
             default_args: dict = dict()
             if issubclass(
                     model_wrapper_type,  # type: ignore