tensorflow.python.framework.errors_impl.NotFoundError: /usr/local/lib/python3.6/dist-packages/tensorflow_text/python/metrics/_text_similarity_metric_ops.so: undefined symbol: _ZN10tensorflow8OpKernel11TraceStringEPNS_15OpKernelContextEb

[Ecclesiast]

Hello!

Could you please help me with this issue? I am using tensorflow image from docker hub tensorflow==2.3.0
My Dockerfile looks like this:

Locally I installed tensorflow==2.3.0 and tensorflow-text==2.3.0 and everything works fine. But when I am using docker image + tensorflow_text I get this issue.

This is my full log:
File "", line 3, in <module> import tensorflow_text File "/usr/local/lib/python3.6/dist-packages/tensorflow_text/__init__.py", line 21, in <module> from tensorflow_text.python import metrics File "/usr/local/lib/python3.6/dist-packages/tensorflow_text/python/metrics/__init__.py", line 20, in <module> from tensorflow_text.python.metrics.text_similarity_metric_ops import * File "/usr/local/lib/python3.6/dist-packages/tensorflow_text/python/metrics/text_similarity_metric_ops.py", line 28, in <module> gen_text_similarity_metric_ops = load_library.load_op_library(resource_loader.get_path_to_datafile('_text_similarity_metric_ops.so')) File "/usr/local/lib/python3.6/dist-packages/tensorflow_core/python/framework/load_library.py", line 61, in load_op_library lib_handle = py_tf.TF_LoadLibrary(library_filename) tensorflow.python.framework.errors_impl.NotFoundError: /usr/local/lib/python3.6/dist-packages/tensorflow_text/python/metrics/_text_similarity_metric_ops.so: undefined symbol: _ZN10tensorflow8OpKernel11TraceStringEPNS_15OpKernelContextEb

[broken]

This error normally shows when TF is has a separate ABI than the TensorFlow that TF.Text was built against. If in this docker image you reinstall TensorFlow using pip (pip install -U tensorflow==2.3.0), do you continue to get this error?

[Ecclesiast]

After installing tensorflow==2.3.0 this issue is gone, thank you!
However, I consider this is a strange logic because I can use container 2.1.0-py3 or 2.3.0 or another container tag and after that install 2.3.0 version.
Intuitively I don't need to install tf again. Could this be fixed in later versions or should I address this issue somewhere else?

[broken]

This is something that should be addressed on TF's side. They are providing you with a binary in this docker image that has a different ABI than what they release publicly. Normally this wouldn't matter, but it does when using libraries that are built to interface with the core TF lib like TF Text.

I expect the docker image did match their build environment at one point, but most likely they updated the build images, but not this other docker image they are providing. After that, they were providing incompatible versions of their library.

Do you mind creating an issue for TF? We've made a lot of progress working with the team to remove inconsistencies from the build environments, but I hadn't run across this one before, and it would be good for them to see this directly affects users. Thanks.

[Ecclesiast]

Thanks for the explanation. Useful to understand.
Yes, I think it's a good idea to create an issue for tensorflow, thanks 👍

[mihaimaruseac]

The docker containers have been republished. Can you check and see if things are working now?

[IrisDinge]

@mihaimaruseac
Nope....
still need do pip install -U tensorflow==2.3.0

[chunyeow]

This issue seems not yet resolved with the following error:

`>>> import tensorflow as tf
Traceback (most recent call last):
File "", line 1, in
File "/home/tensorflow/.local/lib/python3.6/site-packages/tensorflow/init.py", line 436, in
_ll.load_library(_main_dir)
File "/home/tensorflow/.local/lib/python3.6/site-packages/tensorflow/python/framework/load_library.py", line 153, in load_library
py_tf.TF_LoadLibrary(lib)
tensorflow.python.framework.errors_impl.NotFoundError: /usr/local/lib/python3.6/dist-packages/tensorflow/core/kernels/libtfkernel_sobol_op.so: undefined symbol: _ZN10tensorflow8OpKernel11TraceStringEPNS_15OpKernelContextEb

Still need to run pip install -U tensorflow==2.3.0 for the docker image.
`

[puraminy]

I installed 2.3.0 but error still exists
...
File "/home/pouramini/miniconda3/lib/python3.7/site-packages/tensorflow_text/init.py",
line 21, in
from tensorflow_text.python import metrics
File "/home/pouramini/miniconda3/lib/python3.7/site-packages/tensorflow_text/python/metric
s/init.py", line 20, in
from tensorflow_text.python.metrics.text_similarity_metric_ops import *
File "/home/pouramini/miniconda3/lib/python3.7/site-packages/tensorflow_text/python/metric
s/text_similarity_metric_ops.py", line 28, in
gen_text_similarity_metric_ops = load_library.load_op_library(resource_loader.get_path_t
o_datafile('_text_similarity_metric_ops.so'))
File "/home/pouramini/miniconda3/lib/python3.7/site-packages/tensorflow_core/python/framew
ork/load_library.py", line 57, in load_op_library
lib_handle = py_tf.TF_LoadLibrary(library_filename)
tensorflow.python.framework.errors_impl.NotFoundError: /home/pouramini/miniconda3/lib/python
3.7/site-packages/tensorflow_text/python/metrics/_text_similarity_metric_ops.so: undefined s
ymbol: _ZN10tensorflow8OpKernel11TraceStringEPNS_15OpKernelContextEb

[DeathKhan]

Having same issue but on Arch running on bare metal. Seems to be an issue wtih the numpy version they are training against.

[puraminy]

I installed 2.3.0 but error still exists
...
File "/home/pouramini/miniconda3/lib/python3.7/site-packages/tensorflow_text/init.py",
line 21, in
from tensorflow_text.python import metrics
File "/home/pouramini/miniconda3/lib/python3.7/site-packages/tensorflow_text/python/metric
s/init.py", line 20, in
from tensorflow_text.python.metrics.text_similarity_metric_ops import *
File "/home/pouramini/miniconda3/lib/python3.7/site-packages/tensorflow_text/python/metric
s/text_similarity_metric_ops.py", line 28, in
gen_text_similarity_metric_ops = load_library.load_op_library(resource_loader.get_path_t
o_datafile('_text_similarity_metric_ops.so'))
File "/home/pouramini/miniconda3/lib/python3.7/site-packages/tensorflow_core/python/framew
ork/load_library.py", line 57, in load_op_library
lib_handle = py_tf.TF_LoadLibrary(library_filename)
tensorflow.python.framework.errors_impl.NotFoundError: /home/pouramini/miniconda3/lib/python
3.7/site-packages/tensorflow_text/python/metrics/_text_similarity_metric_ops.so: undefined s
ymbol: _ZN10tensorflow8OpKernel11TraceStringEPNS_15OpKernelContextEb

My problem was resolved. I've read that tensorflow and tensorflow_text must be the same version and particularly I've read they should be version 2.3.0 or probably higher. Moreover, my problem could be due conflicts of conda installation and pip installation. so, you'd better to install them both with pip (cause tensorflow_text) isn't on conda) and probably in base environment...

[xmsec]

Having same issue and solution above maybe not useful. My problem could be due to tensorflow that was installed with conda. So i try to pip uninstall tensorflow and reinstall it. Then My problem was resolved.