Op type not registered 'RegexSplitWithOffsets'

[HichemMaiza]

Environment

OS:

Ubuntu : Release 18.04
python : 3.6.9

Dependencies versions:

tensorflow -- 2.0.0
tensorflow-text -- 2.0.1

Error :

2019-12-16 14:59:30.379512: W external/org_tensorflow/tensorflow/core/framework/op_kernel.cc:1622] OP_REQUIRES failed at partitioned_function_ops.cc:113 : Not found: Op type not registered 'RegexSplitWithOffsets' in binary running on pc. Make sure the Op and Kernel are registered in the binary running in this process. Note that if you are loading a saved graph which used ops from tf.contrib, accessing (e.g.) `tf.contrib.resampler` should be done before importing the graph, as contrib ops are lazily registered when the module is first accessed.

Error and Pipeline Description

I am using BertTokenizer in a tensorflow.keras.layers.Layer custom class

Training model.fit --> works fine
Evaluating model.evaluate--> works fine
Saving model.save --> works fine
loading using model.load --> works fine
serving with tensorflow serving --> ERROR

When using tensorflow serving I get the Not found: Op type not registered 'RegexSplitWithOffsets' error

[broken]

This is an issue with the version of TF Text on Tensorflow Serving.
https://github.com/tensorflow/serving/blob/master/tensorflow_serving/workspace.bzl#L64

As mentioned in #173, I sent code to have this updated, but b/c of the holidays it has been slow to get in. I expect this to be updated soon.

[broken]

It took a long time to get the code approved, but the relevant code should now be in tf serving, so this should no longer be an issue.

[cloudy-sfu]

Install tensorflow-text
pip install tensorflow-text --no-dependencies

[HarborZeng]

Install tensorflow-text
pip install tensorflow-text --no-dependencies

And also need to import to register before using:

import tensorflow_text as text  # Registers the ops.