trellixvulnteam / rasa_nlu_zj4r

High talk nalture language understanding platform of LingLing technology . Ltd

HT_NLU

High talk nalture language understanding platform of LingLing technology . Ltd

LingLing Technology Dalian Team's project

Sering bert-as-service model

1、clone HT_NLU project to the local machine

git clone https://github.com/LLGOVRDML/Rasa_NLU.git

2、Serving bert as service model as service
cmd --> Windows+R

pip install bert-serving-server -i https://mirrors.aliyun.com/pypi/simple
pip install bert-serving-client -i https://mirrors.aliyun.com/pypi/simple
cd ${yourpath}/HT_NLU/bert-as-service
bert-serving-start -model_dir D:\chinese_L-12_H-768_A-12 -tuned_model_dir C:\Users\weizhen\Desktop\NLU\rasa_model_output -ckpt_name=model.ckpt-1028

when you see the log print "ready and listening" it means that the bert server is ready , and we can go to the next step

Start rasa_nlu_gq server for classification process

0、cd the root project folder , and double click "visualcppbuildtools full.exe" to install the c++ compiler in windows

1、install related python packages

pip install -r requirements.txt -i https://mirrors.aliyun.com/pypi/simple

2、open project in pycharm and edit execution path
after the previous step , you can open the rasa_nlu_gq project in the pycharm ide
and edit the configuration

in pycharm edit configuration

serving paramaters :

-c sample_configs/config_embedding_bert_intent_estimator_classifier.yml --path projects/bert_gongan_v4

training parameters:

-c sample_configs/config_embedding_bert_intent_estimator_classifier.yml --data data/examples/luis/HighTalkSQSWLuisAppStaging-GA-20180824.json --path projects

3、start the rasa_nlu_gq server
and press the run button for running

test via browser

train rasa process

train rasa nlu with the bert words vectors

python train.py -c sample_configs/config_embedding_bert_intent_classifier.yml --data data/examples/luis/HighTalkSQSWLuisAppStaging-GA-20180824.json --path projects/bert_gongan_v4

Deployed in docker

build rasa docker image

cd rasa_nlu_gq
docker build -t rasa_nlu_gq:v1.0 .

build bert-as-service docker image

cd bert-as-service
docker build -t bert-as-service:v1.0 .

first run bert-as-service image

docker run -it -p 5555:5555 -p 5556:5556 bert-as-service:v1.0

changing rasa_nlu_gq model's ip endpoint and rasa project's ip call out endpoint , after that you can run the rasa docker image

docker run -it -p 5000:5000 rasa_nlu_gq:v1.0