Skip to content

Latest commit

 

History

History
102 lines (75 loc) · 3.37 KB

README.md

File metadata and controls

102 lines (75 loc) · 3.37 KB

Multilingual Machine Translation

This project focuses on developing machine translation models that translate multiple languages into English. We explored various architectures and approaches for this purpose, starting with encoder-decoder models with attention, experimenting with transformers from scratch, and finally leveraging a pretrained model for language detection and translation. The deployment of different models was carried out using FastAPI and Streamlit.

Approach and Methods

  1. Custom Model Development:

    • Encoder-Decoder Architecture:
      We experimented with building custom models using encoder-decoder architectures.
    • Attention Mechanism:
      We incorporated attention mechanisms to improve the model's ability to focus on relevant parts of the input sequence during translation.
    • Languages:
      We focused on Italian, Arabic, Japanese, and French for our initial experiments.

  2. Transformer-Based Model:

    • We developed a transformer-based model specifically for translating Portuguese into English. This required handling tokenization, positional encodings, and multi-head self-attention mechanisms manually.

  3. Pretrained Model (m2m100_418M):

    • For the final phase, we switched to the facebook/m2m100_418M pretrained model, which can handle language detection and translation across multiple languages.
    • Language Detection:
      We utilized the pretrained m2m100_418M model to detect the source language of the input text.
    • Translation:
      Once the language was identified, we employed the model to translate the text into English.
    • Streamlit Deployment:
      We deployed this solution using Streamlit for a user-friendly web interface.

Model Deployment

FastAPI Deployment

  • Used for the initial encoder-decoder and transformer models.
  • Lightweight APIs exposed endpoints for translation.
  • Supported input in Italian, Arabic, Japanese, French, and Portuguese.

Streamlit Deployment

  • Used for the facebook/m2m100_418M model.
  • Interactive web interface for users to input text in any language and view translations.
  • Language detection and translation handled automatically.

Setup and Installation

Prerequisites

  • Python 3.8+
  • Pip package manager
  • Git

Clone the Repository

git clone https://github.com/yourusername/Multilingual_Machine_Translation.git
cd Multilingual_Machine_Translation

Install Dependencies

pip install -r requirements.txt

FastAPI Setup

uvicorn main:app --reload

Streamlit Setup

streamlit run translator_app.py

Usage

  1. FastAPI:

    • Run the FastAPI server using the uvicorn command.

    • Access the API via: http://127.0.0.1:8000/translate

    • Example Request:

      {
  "text": "Olá, como vai?"
}

  2. Streamlit:

    • Run the Streamlit app with the command above.
    • Enter text in the input box, and the translation will appear along with detected language information.

License

This project is licensed under the MIT License - see the LICENSE file for details.

This markdown version maintains the structure and formatting of the original content. You can copy and paste it directly into your `README.md` file.