NER handler

Application architecture:

The application consists of 2 services communicating via REST API:

Main backend or pipeliner that describes the logic of sending requests from one model to another / to the model of the language of choice
NLP service with a model that contains everything for a model to be deployed locally / in GCP

Supported languages:

English

Possible input formats:

Text data as a json

{
    "text_or_file_object": "input text or file object",
    "language": "en",
    "filename": "name of a file"
}

Usage

Each service is packed into a docker container and all of them are managed by docker compose.
To run the application in GCP you must have these secret files and put them as described below in folders structure

creds.json as google credentials

Project structure

Your project structure must look like this:

typical_ner
├── docker-compose_template.yaml
├── creds.json
├── models (if local run ahead)
│   ├──  classification_subheading_m.pt
│   ├──  classification_subheading_n.pt
├── pipeliner
├── model_app

Run application with docker-compose

Install docker-compose (e.g. pip install docker-compose)
Place latest versions of models as described below in models folder

To run docker-compose use:

sudo docker-compose up --build