BERT as a service

There are multiple ways of leveraging the open source BERT model for your NLP work, for example, via huggingface transformers or spacy transformers. I recently came across a post about running BERT as a service, and it was quite easy to setup too.

If the pipeline requires efficient extraction of BERT features this easy setup may save your dev/test time.

Create a environment if needed:

conda create --name bert

Install tensorflow (works with cpu or gpu)

pip3 install tensorflow-cpu==1.15 # version must be before 2.0

OR

pip3 install tensorflow-gpu==1.15 # version must be before 2.0

Install bert serving server and client

pip3 install -U bert-serving-server bert-serving-client

Start the model serving

bert-serving-start -model_dir /path_to_the_model/ -num_worker=2

For example, I used:

bert-serving-start -model_dir ../BERT_models/wwm_uncased_L-24_H-1024_A-16/   -num_worker=4

In Python 3, start using by setting up a client

from bert_serving.client import BertClient

client = BertClient()

vector = client.encode(['lovely portrait'])

References:

• https://towardsdatascience.com/word-embedding-using-bert-in-python-dd5a86c00342
• https://github.com/hanxiao/bert-as-service