Add validation_split support for backend native tensors.

keras/trainers/data_adapters/data_adapter_utils.py

@@ -20,7 +20,7 @@
 if backend.backend() == "tensorflow":
     from keras.utils.module_utils import tensorflow as tf
 
-    ARRAY_TYPES = ARRAY_TYPES + (tf.Tensor, tf.RaggedTensor)
+    ARRAY_TYPES = ARRAY_TYPES + (tf.RaggedTensor,)
 if pandas:
     ARRAY_TYPES = ARRAY_TYPES + (pandas.Series, pandas.DataFrame)
 
@@ -158,7 +158,7 @@ def train_validation_split(arrays, validation_split):
     """
 
     def _can_split(t):
-        return isinstance(t, ARRAY_TYPES) or t is None
+        return backend.is_tensor(t) or isinstance(t, ARRAY_TYPES) or t is None
 
     flat_arrays = tree.flatten(arrays)
     unsplitable = [type(t) for t in flat_arrays if not _can_split(t)]

keras/trainers/trainer_test.py

@@ -291,6 +291,21 @@ def test_fit_with_val_split(
         self.assertIn("loss", history)
         self.assertIn("val_loss", history)
 
+        # Test with backend-native tensors.
+        x = ops.ones((dataset_size, 4))
+        y = ops.zeros((dataset_size, 3))
+        history = model.fit(
+            x,
+            y,
+            batch_size=batch_size,
+            steps_per_epoch=steps_per_epoch if use_steps_per_epoch else None,
+            epochs=epochs,
+            validation_split=0.2,
+        )
+        history = history.history
+        self.assertIn("loss", history)
+        self.assertIn("val_loss", history)
+
     @parameterized.named_parameters(
         [
             ("eager_tf_sparse", True, False),