Segmentation Fault (SIGSEGV) when using tensorflow.keras.config.enable_unsafe_deserialization()

[Pablo-Caballero-Maria]

System information.

• Have I written custom code (as opposed to using a stock example script provided in Keras): No
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Arch Linux 6.13.2
• TensorFlow installed from (source or binary): binary
• TensorFlow version (use command below): 2.18.0
• Python version: 3.12.9
• Bazel version (if compiling from source):
• GPU model and memory: Nvidia MX 130 (2Gb)
• Exact command to reproduce:

You can collect some of this information using our environment capture script:

https://github.com/tensorflow/tensorflow/tree/master/tools/tf_env_collect.sh

You can obtain the TensorFlow version with:
python -c "import tensorflow as tf; print(tf.version.GIT_VERSION, tf.version.VERSION)"

Describe the problem.

When calling tensorflow.keras.models.load_model() after enabling tensorflow.keras.config.enable_unsafe_deserialization(), the script terminates with a segmentation fault (SIGSEGV). Additionally, the logs display sem_destroy: Success before the crash.
The reason why I post the issue in this repository is because, in order to use enable_unsafe_deserialization() from tensorflow.keras, I had to install tf-keras, so I thought the issue arises from here.

Describe the current behavior.

The script stops running and shows this log:

sem_destroy: Success
fish: Job 1, 'python prueba.py' terminated by signal SIGSEGV (Address boundary error)

Describe the expected behavior.

TensorFlow should safely load the model without causing a segmentation fault.

Contributing.

• Do you want to contribute a PR? (yes/no): no
• If yes, please read this page for instructions
• Briefly describe your candidate solution(if contributing):

Standalone code to reproduce the issue.

from tensorflow.keras.models import load_model
from tensorflow.keras import config
config.enable_unsafe_deserialization()
model = load_model('model.keras')

Source code / logs.

Running sudo dmesg | grep -i segfault, I get:

[ 6531.787282] python[89460]: segfault at 60 ip 00007863e831e4d9 sp 00007fff584a0090 error 4 in libpython3.12.so.1.0[31e4d9,7863e807c000+2ad000] likely on CPU 1 (core 1, socket 0)

[tilakrayal]

@Pablo-Caballero-Maria,
With the tensorflow v2.18, the default keras version is 3.0 which can be imported using the command import keras as keras and also you can directly import the config and load using from keras.models import load_model and from keras import config.

Could you please try to use the above commands and test the code. Thank you!