QASystem
this is a QASystem implemented with BERT
install prerequisite packages
install with command
pip3 install -U tf-nightly-2.0-preview bert-for-tf2 flask flask-cors flask-socketio celery gevent
sudo apt install libboost-all-dev rabbitmq-server
celery and socketio need rabbit message queue, so launch it with command
sudo systemctl start rabbitmq-server
download pretrained bert model for Chinese language
download with the following command.
bash downloads.sh
collect question and answer pairs
put the questions and answers in format as question_answer.txt's. and execute following command to convert the collected samples into dataset format.
make -C cc && make -C cc install
./cc/create_dataset -i question_answer.txt -o dataset
finetune bert classifer to predict relavance between question and answer correctly
with directory dataset generated by the above command presented, finetune with the following to start finetune.
python3 Predictor.py
run the QASystem server
run the QA system server by
CUDA_VISIBLE_DEVICES='' python3 server.py
stop the server by Ctrl+C
test the server
firefox <ip>:5000