GramAI

Disclaimer:
This project is developed as a prerequisite to pass the course 02476 Machine Learning Operations taught at the Technical University of Denmark (DTU) in January 2024. All content below is a filled-in template for evaluation purposes. You may take inspiration from how MLOps methodology is applied to develop and maintain an AI-based web app.

Overall goal of the project

In this project, we aim to fine-tune a text2text model to do grammar correction on English sentences. By applying the MLOps methods and tools taught in the course, we want to create a ML project with a focus on maintainability, shareability, and reproducibility.

What framework are you going to use and do you intend to include the framework into your project?

We intend to use the transformer framework from HuggingFace. The framework provides pre-trained models that we can fine tune for our specific use-case.

What data are you going to run on (initially, may change)

We will train our models on a subset of C4 200M Grammar Error Correction dataset (link). The reason for training only a subset is to limit the training time of the models and simplify the scope of the project.

What models do you expect to use

The idea is to use T5. T5 is based on the Transformer architecture, which is a type of neural network that has been proven to be highly effective in NLP tasks. T5 uses a Text-to-Text approach. It is pre-trained on a large text corpus (C4). This allows T5 to learn general language knowledge and transfer it to different tasks. T5 is available in different sizes, ranging from small (60 million parameters) to XXL (11 billion parameters). Our task is to further fine-tune T5 on grammar error correction to improve its performance.

Project structure

The directory structure of the project looks like this:

├── Makefile                  <- Makefile with convenience commands like `make data` or `make train`
├── README.md                 <- The top-level README for developers using this project.
├── data
│   └── processed             <- The final, canonical data sets for modeling.
│
├── models                    <- Trained and serialized models, model predictions, or model summaries
│
├── pyproject.toml            <- Project configuration file
│
├── reports                   <- Generated analysis as HTML, PDF, LaTeX, etc.
│   └── figures               <- Generated graphics and figures to be used in reporting
│
├── dockerfiles               <- Docker files for deployment and training
|
├── config                    <- Config files
|
├── requirements.txt          <- The requirements file for reproducing the analysis environment
|
├── requirements_app.txt      <- The requirements file for deploying the FastAPI based app
|
├── requirements_gapp.txt     <- The requirements file for deploying the graphical based app
|
├── requirements_tests.txt    <- The requirements file for performing tests
|
├── requirements_dev.txt      <- The requirements file for reproducing the analysis environment
│
├── tests                     <- Test files
│
├── cloudbuild.yaml           <- YAML file with steps for using GCP services
│
├── gramai_app_service.yaml   <- YAML file with instructions to replace a service
│
├── src  <- Source code for use in this project.
│   │
│   ├── __init__.py           <- Makes folder a Python module
│   │
│   ├── data                  <- Scripts to download or generate data
│   │   ├── __init__.py
│   │   └── make_dataset.py
│   │
│   ├── train_model.py        <- Script for training the model
│   ├── gramai_app.py         <- Script for deploying app with FastAPI
│   ├── gramai_gapp.py        <- Script for deploying app with Dash to produce a graphical interface for users
│   ├── README.md             <- The top-level README for developers using this project.
│   └── assets
│       │
│       └── style.css         <- CSS file to styling the user graphical app
│
└── LICENSE                   <- Open-source license if one is chosen