TF 2.12 breaks my CNN model

TensorFlow General Discussion
1

I have a convolutional neural network that I have successfully used for training models over the past two years. I upgraded to Tensorflow 2.12 (python3 -m pip install -U tensorflow) a couple weeks ago and my model errors out when I call the fit method. I’ve been looking for a solution for two weeks now, but I still cannot get it to work. Note that when I switch back to version 2.10, my code runs without any issues.

My error follows: You must feed a value for placeholder tensor ‘Placeholder/_0’ with dtype int64 and shape (some_shape), and UNIMPLEMENTED: DNN library is not found.

Any suggestions on how to fix this would be greatly appreciated. More details are available upon request.

3

@Bill,

Could you please include the following code before you importing tensorflow?

import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'

If the issue still exist please share details of your operating system and standalone code to debug the issue?

Thank you!

5

Thanks for the quick response. I made the suggested change and the problem still exists. I’m unable to share my code due to the classified nature of my work. I will try to put a toy problem together next week with fake data that still causes the crash. At that point, I’ll be able to share more details. Thanks.

6

@Bill,

Okay. Please share the data and toy example once you are ready.

Thank you!