Coptic Machine Translator

A Python research module containing code to generate and analyze neural machine translators, parse Coptic datasets, and code to create a simple Google-translate-like frontend. This repo was used to build https://www.coptictranslator.com.

Using the models locally

The best performing models are open-source and have easy to use pipeline APIs through HuggingFace. Refer to the documentation on the model cards for how to run them.

Coptic to English Translator

English to Coptic Translator

Installing the Package

To install the translation package, you must first ensure that your virtual environment has all the required dependencies. Run pip install -e . from the base directory and install all required modules into your virtual environment.

Running the website

Refer to the frontend repo for instrucitons on running the UI locally.

Running this middleware server with a GCP server Endpoint

1. Create a .env file in the server folder
2. Put the following fields in the .env file. Replace <field> with your values.

   COPTIC_TO_ENGLISH_ENDPOINT_ID="<cop-eng endpoint id>"
   ENGLISH_TO_COPTIC_ENDPOINT_ID="<eng-cop endpoint id>"
   PROJECT_ID="<project id>"
   LOCATION="<location>"
   PORT="<port to run the middleware on>"

3. Call

   npm start

Dockerizing a model

To deploy a model, you first need to put in a docker container.

1. Place your model artifacts in models/<model name> folder. This should include the following files
   1. Model config files
   2. Tokenizer files
   3. config.properties - see the documentation here.
   4. handler.py - see the documentation here
   5. pytorch_model.bin - weights
2. Run the docker build script. Use -gpu flag if you plan on serving your model on a GPU. You must include the -m flag with the name of your model in the models/ directory Example:

   ./build_docker.sh -gpu -m cop-en-norm-group-greekified

To run your docker contiainer use

docker run -p 7080:7080 us-central1-docker.pkg.dev/coptic-translation/nmt-docker-repo/<model name>:<processor>

Run this to push your model to the GCP repo (if you have access)

docker push us-central1-docker.pkg.dev/coptic-translation/nmt-docker-repo/<model name>:<processor>

License

This project is licensed under the MIT License - see the LICENSE.md file for details.

MIT License Summary

You can use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the software. This software is provided "AS IS", without warranty of any kind. You must include the original copyright and permission notice in any copy of the software or substantial portion of it. For the full MIT License text, please see the LICENSE.md file in this repository.