Closed msolalh closed 1 year ago
I think the issue is that tensorflow is changing. Is there a way to decide which version to use ? Which one is compatible with the code in the course? THanks.
Are you following the same code that is being written in this repo for food101 milestone project ? If yes then you can downgrade to tensorflow==2.7. I tried it and it gives me no error.
If your code is different then kindly post your steps to reproduce your issue.
The code for food101 milestone project does not work. There is an issue loading EfficientNetB0. I had to change to:
base_model = tf.keras.applications.efficientnet.EfficientNetB0(include_top=False)
But I have a different issue now. Is the code working for you? Thanks, Marc
TypeError Traceback (most recent call last)
4 frames /usr/local/lib/python3.8/dist-packages/keras/utils/traceback_utils.py in error_handler(*args, **kwargs) 65 except Exception as e: # pylint: disable=broad-except 66 filtered_tb = _process_traceback_frames(e.traceback) ---> 67 raise e.with_traceback(filtered_tb) from None 68 finally: 69 del filtered_tb
TypeError: Exception encountered when calling layer "tf.math.truediv_5" (type TFOpLambda).
x
and y
must have the same dtype, got tf.float16 != tf.float32.
Call arguments received by layer "tf.math.truediv_5" (type TFOpLambda): • x=tf.Tensor(shape=(None, None, None, 3), dtype=float16) • y=tf.Tensor(shape=(3,), dtype=float32) • name=None
@msolalh You need to downgrade the tensorflow version to 2.8 then restart your runtime.
Before compiling your model you can check what version of tensorflow you're running on,
tf.__version__
It must output 2.8.0 if you've succesfully downgraded the version and restarted the runtime.
How do downgrade the version? Thanks, Marc
De : Saket Munda @.> Envoyé : jeudi 2 février 2023 03:19 À : mrdbourke/tensorflow-deep-learning @.> Cc : msolalh @.>; Author @.> Objet : Re: [mrdbourke/tensorflow-deep-learning] Food 101 training problem library missing (Issue #504)
You need to downgrade the tensorflow version to 2.8 then restart your runtime. Before compiling your model you can check what version of tensorflow you're running on, tf.version It must output 2.8.0 if you've succesfully downgraded the version and restarted the runtime.
— Reply to this email directly, view it on GitHubhttps://github.com/mrdbourke/tensorflow-deep-learning/issues/504#issuecomment-1413319956, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AXMFS5BPJAZYUXIAZHCGIDLWVNUYNANCNFSM6AAAAAAUCVCHZI. You are receiving this because you authored the thread.Message ID: @.***>
Marc, You can try to run below command in a new cell of a colab,
!pip install tensorflow==2.8
It should start downloading the tensorflow version 2.8
Then when it's done, Restart your Runtime from the Menu Bar,
And you successfully Restarted the Runtime, run,
import tensorflow as tf
print(tf.__version__)
It should print as 2.8.0
It is working now. It looks tensorflow is changing faster than I am doing this course...Any idea why it is not working with current version of tf? Thanks
I think it is working with 2.11 version, but while saving the model after training produces some issues. It could be also because of combination of mixed precision and EfficientNetBX models is a reason. I'm also new to this thing so I haven't dig deeper much to investigate further, but If I find something I'll post my findings here.
@msolalh Also I would like to ask you to tag this issue as answered if you're satisfied with the above answers/discussions. Thanks
Thanks to SaketMunda. This is a tf version issue.
I have issues trying to train for the millestone project. Colab is complaining about a missing DNN library see below. I looked online. It looks something changed in tensorflow new versions. It looks the libraries change faster than I am following this course... Any idea how to fix that problem? Thanks, Marc
Node: 'model_1/efficientnetb0/stem_conv/Conv2D' DNN library is not found. [[{{node model_1/efficientnetb0/stem_conv/Conv2D}}]] [Op:__inference_train_function_24154]