hinglishly-repo

Hinglishly Tanya

This project trains BERT and Seq2Seq models on a Hinglish dataset.

Setup

1. Create a virtual environment:

   python3 -m venv .venv
   source .venv/bin/activate
   

2. Install dependencies:

   pip install -r requirements.txt
   

Data

The dataset is located in the data/ directory:

• train.tsv
• validation.tsv
• test.tsv

Training

Open project.ipynb in VS Code or Jupyter Lab to run the training pipeline.

The notebook covers:

1. Error Detection: Using bert-base-multilingual-cased to classify errors (e.g., Grammar, Spelling, Slang).
2. Grammar Correction: Using google/mt5-small to correct Hinglish sentences.

UI Application

To run the Grammarly-like UI:

streamlit run app.py

Models

The trained models will be saved in:

• bert-domain-classifier/
• mt5-hinglish-correction/

hinglishly-repo