OS: Ubuntu 24.04.2 LTS
Python Version: 3.12.12
Keras Version: 3.11.3
TF Version: 2.20.0

Description

When calling keras.ops.quantile with arguments that results in output dimension > 1, the call results in a ValueError instead of completing successfully.

Reproduction

import os os.environ["KERAS_BACKEND"] = "tensorflow" import numpy as np import tensorflow as tf from keras import ops @tf.function def run_quantile(): x = np.array([[1, 2, 3], [4, 5, 6]]) q = [0.5] return ops.quantile(x, q, axis=1) result = run_quantile() expected_output = np.array([[2, 5]]) assert(np.allclose(expected_output, result))

Expected Result

Assert passes, program terminates without exceptions.

Actual Result

Exception is raised in the call to ops.quantile:
ValueError: Tried to convert 'perm' to a tensor and failed. Error: Expected values deque([1, 0]) to be a dense tensor with shape [], but got shape [2].

Root Cause

When in graph mode, python arguments are automatically converted to Tensorflow tensors. The final transpose call in the internal _quantile function in keras.ops passes a deque to the perm argument. This fails to be converted to a tensor causing the crash.