<aside> <img src="/icons/burst_gray.svg" alt="/icons/burst_gray.svg" width="40px" />

Domains: NLP, Deep Learning, Transformers, Model Deployment, Pre-processing

</aside>

https://github.com/sarayusapa/t5_grammarator

Overview

This project focuses on fine-tuning a transformer model: T5 (Text-to-Text Transfer Transformer) to perform Grammar Error Correction (GEC). The end goal being the model modifying grammatically incorrect sentences into their correct forms.

My Movie 10.mp4

Key Concepts

Neural Network A model inspired by the human brain, that processes complex data by using layers of interconnected artificial “neurons”

Deep Learning A subfield of Machine Learning using large neural networks to perform complex tasks.

Transformer A neural network architecture that processes sequences (like text) and is able to understand context.

Fine-Tuning Training a pretrained model further on a smaller, task-specific dataset to specialize it.

Approach and Workflow

1. Dataset Preprocessing Collect incorrect, correct pairs of sentences as the dataset and format them for training the model.
2. Fine-Tuning Train the T5 transformer on the prepared dataset for the task of grammar correction.
3. Inference and Evaluation Test and measure the performance of this model using different metrics
4. Deployment Make the model available through a FastAPI service, allowing users to send in text (individually or in batches) and get corrected sentences back.