Kernel error on Google Colab

[rahulseetharaman]

Hi, thanks for the implementation in this repo. To give some context, I am trying to run a custom encoder decoder model using this repo and Huggingface transformers.

Steps to reproduce the issue:

1. Clone the repo on colab
2. Install modules using pip install -r requirements.txt
3. When I try to download the model using tf_hub, as follows

# a workaround to load from pytorch checkpoint
from transformers import EncoderDecoderModel, TFEncoderDecoderModel
import tensorflow_hub as tfhub
import tf_sentencepiece
module = tfhub.Module("https://github.com/davidalami/ConveRT/releases/download/1.0/multicontext_tf_model.tar")

I am getting the following error

/usr/local/lib/python3.9/dist-packages/tf_sentencepiece/sentencepiece_processor_ops.py:42: UserWarning: No so file is found for [2.8.0] from [1.15.2, 2.1.0, 2.0.1, 2.2.0, 2.0.0, 1.15.0]
  warnings.warn('No so file is found for [%s] from [%s]' %
/usr/local/lib/python3.9/dist-packages/tf_sentencepiece/sentencepiece_processor_ops.py:44: UserWarning: use the latest version 2.2.0
  warnings.warn('use the latest version %s' % (latest))
---------------------------------------------------------------------------
NotFoundError                             Traceback (most recent call last)
[<ipython-input-1-7224611a5db0>](https://localhost:8080/#) in <cell line: 4>()
      2 from transformers import EncoderDecoderModel, TFEncoderDecoderModel
      3 import tensorflow_hub as tfhub
----> 4 import tf_sentencepiece
      5 
      6 module = tfhub.Module("https://github.com/davidalami/ConveRT/releases/download/1.0/multicontext_tf_model.tar")

2 frames
[/usr/local/lib/python3.9/dist-packages/tensorflow/python/framework/load_library.py](https://localhost:8080/#) in load_op_library(library_filename)
     52     RuntimeError: when unable to load the library or get the python wrappers.
     53   """
---> 54   lib_handle = py_tf.TF_LoadLibrary(library_filename)
     55   try:
     56     wrappers = _pywrap_python_op_gen.GetPythonWrappers(

NotFoundError: /usr/local/lib/python3.9/dist-packages/tf_sentencepiece/_sentencepiece_processor_ops.so.2.2.0: undefined symbol: _ZN10tensorflow8OpKernel11TraceStringEPNS_15OpKernelContextEb

My objective is to use this conveRT module to be fed as encoder into a TFEncoderDecoder model.

Would be grateful for any help.

[davidalami]

Hello @rahulsee , thanks for the issue!
There are a couple of points to mention - the model is trained on tensorflow 1.x, therefore we need to run it in compatibility mode, disabling eager execution. Also, the module requires tensorflow_text to be imported before it is loaded.
The following code works alright:

import tensorflow_hub as tfhub
import tensorflow_text 
import tensorflow as tf

tf.compat.v1.disable_eager_execution()

module = tfhub.Module("https://github.com/davidalami/ConveRT/releases/download/1.0/multicontext_tf_model.tar")

To get more context on how to use the module, please refer here.
Please let me know if there are any other questions :)