SVD operation failed when compute_uv=False (TF backend)

[cofri]

Problem raised

TensorFlow tf.linalg.svd() with compute_uv=False returns a single tensor containing the singular values, and not three tensors "s, u, v" when compute_uv=True.
However, code line 188 in keras/src/backend/tensorflow/linalg.py always expects three returned tensors, even if only one is expected when compute_uv=False. It raises a ValueError: not enough values to unpack.

keras/keras/src/backend/tensorflow/linalg.py

Line 188 in 6abd865

s, u, v = tf.linalg.svd(

)

Tested versions:

• Keras 3.4.1
• TensorFlow 2.17.0

Code to reproduce:

import keras
keras.ops.svd([[1.0, 0.0], [2.0, 0.0]], compute_uv=False)

[google-ml-butler]

Are you satisfied with the resolution of your issue?
Yes
No