DA6401 - Assignment 3: Implementing the Transformer for Machine Translation

Overview

In this assignment, you will implement the landmark architecture from the paper "Attention Is All You Need" from scratch using PyTorch. The goal is to develop a Neural Machine Translation (NMT) system capable of translating text from German to English using the Multi30k dataset.

Project Structure

assignment3/
├── requirements.txt
├── README.md
├── model.py           # Core Transformer architecture (Encoders, Decoders, Multi-Head Attention)
├── utils.py           # Label Smoothing, Noam Scheduler, Masking Utilities
├── dataset.py         # Multi30k dataset loading and spacy tokenization
├── train.py           # Training loops and Greedy Decoding inference