.
├── Dockerfile                              # API service on containers
├── Jenkinsfile                             # Jenkins file used for automation
├── README.md
├── requirements.txt                        # required libraries
├── startup_api.sh                          # start API service
├── ..........
└── textMG                                  # source directory
    ├── APIs                                # API codes
    │   ├── api_logger                      # API logger
    │   ├── base_models                     # pydantic BaseModel
    │   ├── db                              # SQLAlchemy table definition
    │   ├── html                            # html files
    │   └── routers                         # API routers
    │       ├── data_generator.py
    │       ├── health_check.py
    │       ├── tensorflow_service.py       # all training/eval/pred/refer APIs 
    │       ├── users.py                    # user management APIs
    │       ├── users_token.py              # get secret keys db and token releated components
    │       └── .....
    ├── configs                             # args config files
    │   ├── config.py
    │   └── config_multiGPU.py
    ├── datasets                            # for tools used for data loading&processing
    │   ├── data_loader.py                  
    │   ├── dataset.py                      # tools for data manipulating
    │   └── generator.py                    # generator for data batch 
    ├── embeddings                          # word2vec model embeddings 
    ├── logs                                # logs file
    ├── models                              # deep learning models
    ├── test                                # testing for each components
    ├── tf_serving                          # tensorflow serving grpc and restful
    ├── utils                               # utils tools
    ├── main_multiGPU.py                    # main entrypoint for model training/eval/pred
    ├── api_run.py                          # main entrypoint for API services 
    ├── startup.sh                          # another entrypoint for model training/eval/pred
    └── ..........

$ pip install -r requirements.txt

python setup.py sdist
pip install TFpackageText-0.0.1.tar.gz

from textMG import models, datasets, APIs

$ ./starup.sh -m [train|evel|pred]

http://localhost:5000/api/v1/models/train
http://localhost:5000/api/v1/models/eval
http://localhost:5000/api/v1/models/pred
http://localhost:5000/api/v1/models/infer
http://localhost:5000/api/v1/models/pro_cons_infer

class Bert_module:
    def __init__(self, *args, **kwargs):
        self.bert = BertModel
        self.kwargs = kwargs

    def __call__(self, num_labels, max_len, hidden_size, reuse=tf.AUTO_REUSE, is_training=True, dropout=0.3, *args, **kwargs):
        with tf.variable_scope('bert_classification', reuse=reuse):
            output_weights = tf.get_variable(
                "output_weights", [num_labels, hidden_size],
                initializer=tf.truncated_normal_initializer(stddev=0.02))

$ python api_run.py

$ docker build -t api_docker .
$ docker run -t -p 80:5000 -v /home/projects/TFpackageText/textMG:/home/projects --name=api_docker_service api_docker

$ gunicorn api_run:app --workers 2 --worker-class uvicorn.workers.UvicornWorker --bind 0.0.0.0:5000 --log-level debug

generate a secret key and save it:
$ openssl rand -hex 32

generate hashed password:
$ from passlib.context import CryptContext
$ hashed_password = CryptContext(schemes=["bcrypt"], deprecated="auto").hash('admin')

create an admin user:
$ insert into users_v(id, first_name,last_name,email,hashed_password) values(1, 'admin', 'v', '[email protected]', '$hashed_password')

add secret key, algothrim .. into db:
$ insert into token_users_v(SECRET_KEY, ALGORITHM, ACCESS_TOKEN_EXPIRE_MINUTES, uid) values('ec039bcb8eada04c0de3cea9ced7440b949f6e2dadafd61ae1f8b687c214b4a9', 'HS256', 30, 1);

authorize yourself with below username/passwd on Swagger docs: http://localhost:5000/docs:
$ username: admin
$ password: admin

generate JWT tokens used in "-H 'Authorization: Bearer JWT tokens'":
$ curl -X 'POST' \
  'http://localhost:5000/token' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/x-www-form-urlencoded' \
  -d 'grant_type=&username=admin&password=admin&scope=&client_id=&client_secret='

http://localhost:5000/docs

# load data from disk and generator.
generator = Generator()
iter = generator.get_next_patch(batch=2)
el = next(iter)

train_dataset = tf.data.Dataset.from_generator(generator.get_next_patch,
                                               output_types={generator.input: tf.float32,
                                                             generator.output: tf.float32})
train_dataset = train_dataset.repeat().batch(args.batch_size * args.num_gpusORcpus).prefetch(1)

curl -X 'GET' \
  'http://localhost:5000/api/models/batch_load/2' \
  -H 'accept: application/json'
  -H 'Authorization: Bearer eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJzdWIiOiJ2aWM3IiwiZXhwIjoxNjU0NzA2NDk0fQ.VPNdR2Ow0vwl4Yx3JYmErSdCZP6G8RufSUAk1lavR2M'

# call build_SavedModel to build a SavedModel for tensor serving
build_SavedModel(conv_net, args.export_path_serving, args.savedmodel_version, X_batch_numGPU, best_sess)

$ docker pull tensorflow-serving:1.15.0
$ nohup docker run --name container_serving -p 8501:8501 -p 8500:8500 \
    ........
    --batching_parameters_file=/models/batch.config >  /home/projects/tf1_serving_log.out 2>&1 &

curl -X 'POST' \
  'http://localhost:5000/api/v1/models/infer' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJzdWIiOiJ2aWM3IiwiZXhwIjoxNjU0NzA2NDk0fQ.VPNdR2Ow0vwl4Yx3JYmErSdCZP6G8RufSUAk1lavR2M' \
  -d '{
  "batch_infer": [
    "string"
  ]
}'

curl -X 'POST' \
  'http://localhost:5000/api/v1/models/pro_cons_infer' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJzdWIiOiJ2aWM3IiwiZXhwIjoxNjU0NzA2NDk0fQ.VPNdR2Ow0vwl4Yx3JYmErSdCZP6G8RufSUAk1lavR2M' \
  -d '{
  "batch_infer": [
    "string"
  ]
}'

node("jekins-agent1"){
    stage("Git clone"){
        sh 'cd /opt/jenkins/workspace/  && rm -rf TFpackegeText'
        sh 'git clone [email protected]:victorbai2/TFpackageText.git && tar -czvf package.tar.gz'
    }
    .......

curl -X 'POST' \
  'http://localhost:5000/api/v1/models/pred' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  -H 'Authorization: Bearer eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJzdWIiOiJ2aWM3IiwiZXhwIjoxNjU0NzA2NDk0fQ.VPNdR2Ow0vwl4Yx3JYmErSdCZP6G8RufSUAk1lavR2M' \
  -d '{
  "inquiry": [
    "在北京举办...", "三点不足导致球队...."...
  ]
}'

Response Body:

{
  "data": {
    "records": {
      "0": {
        "category": "entertainment",
        "label": "1",
        "pred_logits": "[0.17890048027038574, 0.23429206013679504, 0.220640629529953, 0.17155364155769348, 0.1946132332086563]"
      },
      "1": {
        "category": "entertainment",
        "label": "1",
        "pred_logits": "[0.17201262712478638, 0.24676887691020966, 0.22940053045749664, 0.15629342198371887, 0.19552461802959442]"
      }
    },
    "time_used": "0.519162654876709"
  },
  "code": 200,
  "message": "success",
  "error": false
}

curl -X 'GET' \
  'http://localhost:5000/api/v1/models/batch_load/2' \
  -H 'accept: application/json' \
  -H 'Authorization: Bearer eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJzdWIiOiJ2aWM3IiwiZXhwIjoxNjU0NzA2NDk0fQ.VPNdR2Ow0vwl4Yx3JYmErSdCZP6G8RufSUAk1lavR2M' \
  
Response Body:
{
  "data": {
    "batch": {
      "x_input": [
        [701,101,..],
        [1842,2182,...]
      ],
      "y_output": [
        [0,0,0,0,1],
        [0,0,0,0,1],
      ]
    },
    "input_shape": [2,30],
    "output_shape": 2
  },
  "code": 200,
  "message": "success",
  "error": false
}

$ ab -p tensorAPI.json  -T 'application/json' -H 'Content-Type: application/json'  \
  -H 'Authorization: Bearer eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJzdWIiOiJ2aWM3IiwiZXhwIjoxNjU0NzA2NDk0fQ.VPNdR2Ow0vwl4Yx3JYmErSdCZP6G8RufSUAk1lavR2M' \
  -c 500 -n 500 http://localhost:5000/api/v1/models/infer

$ ab -p tensorAPI.json  -T 'application/json' -H 'Content-Type: application/json'  \
  -H 'Authorization: Bearer eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJzdWIiOiJ2aWM3IiwiZXhwIjoxNjU0NzA2NDk0fQ.VPNdR2Ow0vwl4Yx3JYmErSdCZP6G8RufSUAk1lavR2M' \
  -c 500 -n 500 http://localhost:5000/api/v1/models/pro_cons_infer