Support scalar view for tf backend. (#21802)

* Support scalar view for tf backend.

* update last dim

Author: wenyi-guo

keras/src/backend/tensorflow/numpy.py

@@ -1010,7 +1010,11 @@ def view(x, dtype=None):
     old_itemsize = old_dtype.size
     new_itemsize = new_dtype.size
 
-    if list(x.shape)[-1] * old_itemsize % new_itemsize != 0:
+    old_shape = list(shape_op(x))
+    last_dim_size = old_shape[-1] if len(old_shape) > 0 else -1
+    if (last_dim_size == -1 and old_itemsize != new_itemsize) or (
+        last_dim_size * old_itemsize % new_itemsize != 0
+    ):
         raise ValueError(
             f"Cannot view array of shape {x.shape} and dtype {old_dtype} "
             f"as dtype {new_dtype} because the total number of bytes "
@@ -1027,8 +1031,6 @@ def view(x, dtype=None):
         cast_tensor = tf.bitcast(flat_tensor, type=new_dtype)
         return tf.reshape(cast_tensor, new_shape)
     else:
-        old_shape = list(shape_op(x))
-        last_dim_size = old_shape[-1]
         ratio = new_itemsize // old_itemsize
         if isinstance(last_dim_size, int) and last_dim_size % ratio != 0:
             raise ValueError(

keras/src/ops/numpy_test.py

@@ -4136,6 +4136,16 @@ def test_concatenate(self):
         )
 
     def test_view(self):
+        x = np.array(1, dtype="int16")
+        result = knp.view(x, dtype="float16")
+        assert backend.standardize_dtype(result.dtype) == "float16"
+
+        with self.assertRaises(Exception):
+            result = knp.view(x, dtype="int8")
+
+        with self.assertRaises(Exception):
+            result = knp.view(x, dtype="int32")
+
         x = np.array([[1, 2, 3, 4], [5, 6, 7, 8]], dtype="int16")
         result = knp.view(x, dtype="int16")
         assert backend.standardize_dtype(result.dtype) == "int16"