diff --git a/examples/distributed_training_with_flexible_runner.py b/examples/distributed_training_with_flexible_runner.py
index 20e7286695..99d2cf257d 100644
--- a/examples/distributed_training_with_flexible_runner.py
+++ b/examples/distributed_training_with_flexible_runner.py
@@ -94,6 +94,9 @@ def main():
                 initial_scale_power=15,
             ),
             inputs_to_half=[0],
+            # bf16=dict(
+            #     enabled=True,
+            # ),
             zero_optimization=dict(
                 stage=3,
                 allgather_partitions=True,
diff --git a/mmengine/_strategy/deepspeed.py b/mmengine/_strategy/deepspeed.py
index 2565d18c7d..a439a4a952 100644
--- a/mmengine/_strategy/deepspeed.py
+++ b/mmengine/_strategy/deepspeed.py
@@ -18,7 +18,7 @@
 from mmengine.optim import BaseOptimWrapper, _ParamScheduler
 from mmengine.registry import (MODEL_WRAPPERS, OPTIM_WRAPPERS, OPTIMIZERS,
                                STRATEGIES)
-from mmengine.utils import digit_version, get_git_hash
+from mmengine.utils import apply_to, digit_version, get_git_hash
 from .base import BaseStrategy
 
 
@@ -188,18 +188,22 @@ def _cast_inputs_half(self, inputs: Union[list, tuple, dict, None]):
         if self._inputs_to_half is None:
             return inputs
 
+        dtype = next(self.model.parameters()).dtype
         if isinstance(inputs, (list, tuple)):
             new_inputs = []
             for i, v in enumerate(inputs):
                 if i in self._inputs_to_half:
-                    new_inputs.append(v.half())
+                    new_inputs.append(
+                        apply_to(v, lambda x: hasattr(x, 'to'),
+                                 lambda x: x.to(dtype)))
                 else:
                     new_inputs.append(v)
             return inputs.__class__(new_inputs)
         elif isinstance(inputs, dict):
             for k, v in inputs.items():
                 if k in self._inputs_to_half:
-                    inputs[k] = v.half()
+                    inputs[k] = apply_to(v, lambda x: hasattr(x, 'to'),
+                                         lambda x: x.to(dtype))
             return inputs
         else:
             raise TypeError('inputs should be list, tuple or dict, '