# English to Hindi Text Translation using Transformers

This project showcases a simple text translation model that translates English text to Hindi using the Hugging Face Transformers library. The model utilizes pre-trained sequence-to-sequence architecture for accurate and efficient translation.

## Table of Contents

- [Project Overview](#project-overview)
- [Installation](#installation)
- [Usage](#usage)
- [Model Training and Dataset](#model-training-and-dataset)
- [Model Testing and Deployment](#model-testing-and-deployment)
- [User Interface](#user-interface)
- [Challenges Faced](#challenges-faced)
- [Contributions](#contributions)

## Project Overview

Text translation is an essential task in natural language processing, and this project aims to provide a practical example of building and deploying a translation model. The project covers the following aspects:

- Data preprocessing: Tokenization and dataset preparation.
- Model training: Training a sequence-to-sequence model for English-to-Hindi translation.
- Model testing: Translating text using the trained model.
- User interface: Creating a user-friendly interface for text translation.

## Installation

To run this project, you'll need the following dependencies:

- Python 3.x
- TensorFlow
- Hugging Face Transformers
- Datasets library
- Gradio

You can install the required libraries using the following shell command:

```shell
pip install datasets transformers[sentencepiece] tensorflow gradio -q
```

## Usage
Checkout the app [here](https://huggingface.co/spaces/Lohith9923/En-Hi-Translation) where you need to give english sentences or text in input textbox and output is translated text or sentence in Hindi. You can see some examples down for checking.

## Model Training and Dataset
For training the text translation model.
You can checkout the pre-trained model from [here](https://colab.research.google.com/corgiredirector?site=https%3A%2F%2Fhuggingface.co%2FHelsinki-NLP%2Fopus-mt-en-hi) and Dataset from [here](https://huggingface.co/datasets/cfilt/iitb-english-hindi/viewer/cfilt--iitb-english-hindi).
- First Download the pre-trained model using **transformers** library in python.
- Load the Dataset **cfilt/iitb-english-hindi** using **Datasets** library in python.
- Initialized the model, tokenizer, and preprocessing function.
- Tokenized the dataset and prepared the training and validation data.
- Compiled the model with the optimizer(**Adam**) with required parameters.
- Trained the model for the desired number of epochs.

## Model Testing and Deployment
To test the trained model and deploy a user interface:

- Saved the trained model at a preferred location.
- Loaded the model from the location and tokenizer for testing.
- Translated sample input text using the model.
- Deployed a Gradio interface for user-friendly translation.

## User Interface

The Gradio interface provides an interactive way to translate English text to Hindi. To use the interface:

- Run the project and navigate to the specified URL.
- Enter English text in the input box.
- Checkout the translated Hindi text in the output box.

## Challenges Faced

- Surfed through lot of resources in google and other platforms for best dataset for my project.
- Spent a lot of time gathering the correct resources for understanding about transformers, LLM's and gradio.

## Contributions
Contributions to this project are welcome! Here are some ways you can contribute:

- Improve the model's translation quality and performance.
- Enhance the user interface for a better user experience.
- Add support for more languages and translation directions.

To contribute, follow these steps:

- Fork this repository.
- Create a new branch for your feature or bug fix.
- Commit your changes and push them to your fork.
- Open a pull request with a detailed description of your changes.