Apply tensorflow 2.0's v1 compatibility layer so the code works with TF2.

[mmohaveri]

The current master does not work with Tensorflow 2.0 because it's using a set of deprecated APIs.
This PR will replace those deprecated APIs with the ones from "tf.compat.v1".

[AndreaSottana]

Hi @mmohaveri
I have tried using your code. However, tf is not able to build the graph correctly, at least for me.
I have downloaded the BER model, I then open a terminal and run the following
bert-serving-start -model_dir=/d/Users/[...]/uncased_L-12_H-768_A-12 -cpu

But I then get the following error

Traceback (most recent call last):
  File "D:\Users\[...]\venv38\Scripts\bert-serving-start-script.py", line 33, in <module>
    sys.exit(load_entry_point('bert-serving-server', 'console_scripts', 'bert-serving-start')())
  File "d:\users\[...]\bert-as-service\server\bert_serving\server\cli\__init__.py", line 8, in main
    with BertServer(get_run_args()) as server:
  File "d:\users\[...]\bert-as-service\server\bert_serving\server\__init__.py", line 75, in __init__
    self.graph_path, self.bert_config = pool.apply(optimize_graph, (self.args,))
TypeError: cannot unpack non-iterable NoneType object

even though the Bert model is there. Any idea what might be going wrong?

[Akila-Ayanthi]

@AndreaSottana were you able to get it to work?

[Akila-Ayanthi]

@mmohaveri How do you run BAS from the source code?

[AndreaSottana]

@Akila-Ayanthi No I didn't, I suspect the original code from hanxiao:master (last commit August 2020) is not
regularly maintained and something might have broken due to version issues.
I am using this very good alternative though, to integrate BERT with ElasticSearch https://github.com/UKPLab/sentence-transformers/blob/master/examples/applications/semantic-search/semantic_search_quora_elasticsearch.py

[Akila-Ayanthi]

@AndreaSottana Thank you.