diff --git a/keras_hub/src/models/task_test.py b/keras_hub/src/models/task_test.py
index bf57c88912..ac4f8def5b 100644
--- a/keras_hub/src/models/task_test.py
+++ b/keras_hub/src/models/task_test.py
@@ -147,6 +147,10 @@ def test_save_to_preset(self):
         new_out = restored_task.backbone.predict(data)
         self.assertAllClose(ref_out, new_out)
 
+        # Check setting dtype.
+        restored_task = TextClassifier.from_preset(save_dir, dtype="float16")
+        self.assertEqual("float16", restored_task.backbone.dtype_policy.name)
+
     @pytest.mark.large
     def test_save_to_preset_custom_backbone_and_preprocessor(self):
         preprocessor = keras.layers.Rescaling(1 / 255.0)
diff --git a/keras_hub/src/utils/preset_utils.py b/keras_hub/src/utils/preset_utils.py
index 52aad373a0..4f80f5227f 100644
--- a/keras_hub/src/utils/preset_utils.py
+++ b/keras_hub/src/utils/preset_utils.py
@@ -658,6 +658,12 @@ def load_task(self, cls, load_weights, load_task_weights, **kwargs):
                 cls, load_weights, load_task_weights, **kwargs
             )
         # We found a `task.json` with a complete config for our class.
+        # Forward backbone args.
+        backbone_kwargs, kwargs = self.get_backbone_kwargs(**kwargs)
+        if "backbone" in task_config["config"]:
+            backbone_config = task_config["config"]["backbone"]["config"]
+            backbone_config = {**backbone_config, **backbone_kwargs}
+            task_config["config"]["backbone"]["config"] = backbone_config
         task = load_serialized_object(task_config, **kwargs)
         if task.preprocessor and hasattr(
             task.preprocessor, "load_preset_assets"