keras.ops.flip fails for eager tensors when axis is a list of axes

[AGFACBNNR]

keras.ops.flip() raises a TensorFlow ReverseV2 error when used with an eager tensor and axis provided as a list of axes, even though the API docs say axis may be an axis or axes.

Reproduction

import keras
import tensorflow as tf

x = tf.ones((3, 4, 5))
print(keras.ops.flip(x, axis=[0, 1]))

Expected behavior

The tensor should be flipped along axes 0 and 1, returning a tensor of the same shape.

Actual behavior

Traceback (most recent call last):
  File "/mnt/data2/Users/sunshuo/2026/dltest_3.3/src/test.py", line 5, in <module>
    print(keras.ops.flip(x, axis=[1]))
          ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/mnt/data2/Users/sunshuo/anaconda3/envs/dltest_to_raise_issue/lib/python3.12/site-packages/keras/src/ops/numpy.py", line 3338, in flip
    return backend.numpy.flip(x, axis=axis)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/mnt/data2/Users/sunshuo/anaconda3/envs/dltest_to_raise_issue/lib/python3.12/site-packages/keras/src/backend/tensorflow/numpy.py", line 1552, in flip
    return tf.reverse(x, [axis])
           ^^^^^^^^^^^^^^^^^^^^^
  File "/mnt/data2/Users/sunshuo/anaconda3/envs/dltest_to_raise_issue/lib/python3.12/site-packages/tensorflow/python/ops/gen_array_ops.py", line 9263, in reverse_v2
    return reverse_v2_eager_fallback(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/mnt/data2/Users/sunshuo/anaconda3/envs/dltest_to_raise_issue/lib/python3.12/site-packages/tensorflow/python/ops/gen_array_ops.py", line 9309, in reverse_v2_eager_fallback
    _result = _execute.execute(b"ReverseV2", 1, inputs=_inputs_flat,
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/mnt/data2/Users/sunshuo/anaconda3/envs/dltest_to_raise_issue/lib/python3.12/site-packages/tensorflow/python/eager/execute.py", line 53, in quick_execute
    tensors = pywrap_tfe.TFE_Py_Execute(ctx._handle, device_name, op_name,
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
tensorflow.python.framework.errors_impl.InvalidArgumentError: {{function_node __wrapped__ReverseV2_device_/job:localhost/replica:0/task:0/device:CPU:0}} 'dims' must be 1-dimension, not 2 [Op:ReverseV2]```

[divyashreepathihalli]

https://github.com/keras-team /triage

[divyashreepathihalli]

@keras-team /triage

[github-actions]

The issue is caused by the implementation of flip in the TensorFlow backend, which incorrectly wraps the axis parameter in a list even if it is already a list or tuple. This results in a nested structure (e.g., [[0, 1]]) that tf.reverse cannot handle, leading to the reported InvalidArgumentError.

Suggested Fix

In keras/src/backend/tensorflow/numpy.py, the flip function should be updated to use the to_tuple_or_list utility (which is already imported in that file) to handle both scalar and list-like axes correctly:

def flip(x, axis=None):
    x = convert_to_tensor(x)
    if axis is None:
        return tf.reverse(x, tf.range(tf.rank(x)))
    axis = to_tuple_or_list(axis)
    return tf.reverse(x, axis)

This change ensures that axis is always a 1D structure when passed to tf.reverse, matching the expected behavior described in the Keras API documentation.