RuntimeError: when unable to load the library or get the python wrappers. about vit-keras

[faustomorales]

Yes, the intention is for this package to work with the latest available version of TensorFlow. Could you provide the following?

• The version of TensorFlow that didn't work
• The error (along with a full traceback)

The dependencies for this package are specified using pyproject.toml, which would make a requirements.txt superfluous. TensorFlow is not specified there in order to allow people to install their preferred flavor of the package (e.g., some people still use tensorflow-gpu or build from source).

[pranavchat14]

I am using Google Colab. I am doing exactly like instructed in README.md.

I have installed vit-keras using pip. Then I go with this example code:

image_size = 224
model = vit.vit_l32(
image_size=image_size,
activation='sigmoid',
pretrained=True,
include_top=True,
pretrained_top=False,
classes=200
)

The details are as follows:

• The version of TensorFlow that didn't work: tensorflow==2.4.1

• The error message: Below

---

NotFoundError Traceback (most recent call last)
in ()
8 include_top=True,
9 pretrained_top=False,
---> 10 classes=200
11 )

14 frames
/usr/local/lib/python3.6/dist-packages/tensorflow/python/framework/load_library.py in load_op_library(library_filename)
55 RuntimeError: when unable to load the library or get the python wrappers.
56 """
---> 57 lib_handle = py_tf.TF_LoadLibrary(library_filename)
58 try:
59 wrappers = _pywrap_python_op_gen.GetPythonWrappers(

NotFoundError: /usr/local/lib/python3.6/dist-packages/tensorflow_addons/custom_ops/activations/_activation_ops.so: undefined symbol: _ZN10tensorflow14kernel_factory17OpKernelRegistrar12InitInternalEPKNS_9KernelDefEN4absl11string_viewESt10unique_ptrINS0_15OpKernelFactoryESt14default_deleteIS8_EE

[faustomorales]

Ah, you've run into a compatibility problem between tensorflow_addons and tensorflow. Unfortunately, users have to manage the version compatibility for these two packages manually. See the tensorflow_addons README for more information on why that's the case. Unfortunately, it's not something that vit-keras can handle automatically.

Because Google Colab periodically updates the version of TensorFlow that ships, it's also not easy to just set a version of tensorflow_addons to fix it. For that reason, I suggest setting a fixed version for both tensorflow and tensorflow_addons at the top of your notebook.

As an example, if you want to use tensorflow==2.4.1, you can look it up on the tensorflow_addons compatibility chart and see that it requires tensorflow_addons==0.12.1. And so the following would work in a Google Colab notebook.

!pip install -q vit-keras tensorflow==2.4.1 tensorflow_addons==0.12.1

import vit_keras.vit as vit

image_size = 224
model = vit.vit_l32(
  image_size=image_size,
  activation='sigmoid',
  pretrained=True,
  include_top=True,
  pretrained_top=False,
  classes=200
)

Closing this for now but let me know if this doesn't solve your problem.