_activation_ops.so: undefined symbol error when importing tensorflow_addons

[martin-gorner]

System information

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Google Cloud AI Platform Notebooks, TF 2.0 image
• TensorFlow version and how it was installed (source or binary): Google Cloud AI Platform Notebooks, TF 2.0 image
• TensorFlow-Addons version and how it was installed (source or binary): 0.6.0, pip install
• Python version: 3.5
• Is GPU used? (yes/no): yes

Executing the following line in my notebook:
import tensorflow_addons as tfa
Generates the following error:
NotFoundError: /home/jupyter/.local/lib/python3.5/site-packages/tensorflow_addons/custom_ops/activations/_activation_ops.so: undefined symbol: _ZN10tensorflow12OpDefBuilder4AttrESs
Full error message:
NotFoundError Traceback (most recent call last)
in
5 from matplotlib import pyplot as plt
6 import PIL.Image, PIL.ImageFont, PIL.ImageDraw
----> 7 import tensorflow_addons as tfa
8 AUTOTUNE = tf.data.experimental.AUTOTUNE
9 print("Tensorflow version " + tf.version)

~/.local/lib/python3.5/site-packages/tensorflow_addons/init.py in
19
20 # Local project imports
---> 21 from tensorflow_addons import activations
22 from tensorflow_addons import callbacks
23 from tensorflow_addons import image

~/.local/lib/python3.5/site-packages/tensorflow_addons/activations/init.py in
19 from future import print_function
20
---> 21 from tensorflow_addons.activations.gelu import gelu
22 from tensorflow_addons.activations.hardshrink import hardshrink
23 from tensorflow_addons.activations.lisht import lisht

~/.local/lib/python3.5/site-packages/tensorflow_addons/activations/gelu.py in
23
24 _activation_ops_so = tf.load_op_library(
---> 25 get_path_to_datafile("custom_ops/activations/_activation_ops.so"))
26
27

/usr/local/lib/python3.5/dist-packages/tensorflow_core/python/framework/load_library.py in load_op_library(library_filename)
59 RuntimeError: when unable to load the library or get the python wrappers.
60 """
---> 61 lib_handle = py_tf.TF_LoadLibrary(library_filename)
62
63 op_list_str = py_tf.TF_GetOpList(lib_handle)

NotFoundError: /home/jupyter/.local/lib/python3.5/site-packages/tensorflow_addons/custom_ops/activations/_activation_ops.so: undefined symbol: _ZN10tensorflow12OpDefBuilder4AttrESs

[seanpmorgan]

Same issue found here:
#676 (comment)

Thanks for reporting, we're aware of the issue and a long term fix is in the works (see RFC linked in that issue). A short term fix would be to try compiling TFA from source (but there are a few edge cases where it could still fail