English to Hindi Neural Machine Translator

This project implements a neural machine translation system for translating English text to Hindi using a sequence-to-sequence (Seq2Seq) architecture with attention mechanism. The model is built with PyTorch and features a GRU-based encoder-decoder network.

Features

• Seq2Seq architecture with attention mechanism
• GRU-based encoder and decoder
• Customizable model parameters (hidden size, embedding dimensions, etc.)
• Interactive translation through command line interface
• Web interface using Gradio
• Support for vocabulary management and data preprocessing
• Teacher forcing during training

Project Structure

├── app.py                 # Gradio web interface for translation
├── inference.py           # Functions for model inference
├── train.py               # Model training script
├── models/                # Neural network architecture components
│   ├── encoder.py         # Encoder implementation
│   ├── decoder.py         # Decoder implementation  
│   ├── attention.py       # Attention mechanism
│   └── seq2seq.py         # Seq2Seq model
├── utils/                 # Utility functions
│   ├── config.py          # Configuration parameters
│   ├── data_loader.py     # Data loading utilities
│   └── preprocessing.py   # Text preprocessing functions
├── data/                  # Data directory
│   └── hindi_english_parallel.csv  # Parallel corpus
└── bin/                   # Model checkpoints and vocabularies
    ├── seq2seq.pth        # Trained model weights
    ├── eng_vocab.pkl      # English vocabulary
    └── hin_vocab.pkl      # Hindi vocabulary

Installation

1. Clone the repository:

git clone https://github.com/yourusername/english-hindi-translator.git
cd english-hindi-translator

2. Install the required packages:

pip install -r requirment.txt

Usage

Training the Model

To train the translation model:

python train.py

This will:

• Load and preprocess the Hindi-English parallel corpus
• Build vocabularies for both languages
• Initialize and train the Seq2Seq model
• Save model checkpoints after each epoch

Translation via Command Line

For interactive translation through command line:

python inference.py

Web Interface

To launch the web interface for translation:

python app.py

This will start a Gradio interface that you can access in your web browser.

Model Architecture

• Encoder: GRU-based with configurable layers and embedding dimensions
• Decoder: GRU with attention mechanism
• Attention: Calculates attention scores between encoder outputs and decoder hidden states
• Training: Uses teacher forcing and cross-entropy loss

Configuration

Model parameters can be adjusted in the config.py file:

• embedding_dim: Size of word embeddings
• hidden_size: Size of hidden layers
• num_layers: Number of RNN layers
• dropout: Dropout rate
• batch_size: Training batch size
• learning_rate: Learning rate for optimizer
• epochs: Number of training epochs
• teacher_forcing_ratio: Ratio of teacher forcing during training
• max_vocab_english: Maximum size of English vocabulary
• max_vocab_hindi: Maximum size of Hindi vocabulary
• max_length: Maximum sentence length

Requirements

• torch
• pandas
• numpy
• tqdm
• gradio