This project is a minor modification to https://github.com/ditadi/ner
A dockerized and deployable Named Entity Recognition tool for Icelandic, using a ELECTRA-base Icelandic language model fined tuned for NER on the MIM-GOLD-NER corpus.
- Make sure that Docker is properly installed
- downlaod the zipped pytorch models directory and extract to directory ./src/bert/
- run the following commands starting in root directory of the project to deploy locally
cd src/bert
make build
make run
All the API calls use post and input/outputs are in a json format. Further details about the api calls are automatically generated when the container is run and can be found in /docs or /redoc
HTTP METHOD | Description |
---|---|
/predict | Takes in Icelandic text and returns text with icelandic named entetiy recognition |
test files can be found in tests/
. There are two tests that can be performed.
- Normal api tests: this is where you test the api from the running docker image
- ELG api tests: this is where you run
docker-compose up
and get an instance as if you where running the docker container on ELG. To submit a api call you then need to call/process/service
.
This ELG API was developed in EU's CEF project: Microservices at your service
The underlying Named Entity Recognizer is NER for AISC ML Ops by dmitriy, which is licensed under this MIT license. The original Icelandic NER API in the master branch of this repository was a minor modification to the underlying tool. Then the project was updated to conform to the ELG API standard. This version of Icelandic NER API is copied into the docker image when it is built.