This repository contains an official TensorFlow tutorial demonstrating how to build neural machine translation systems from scratch. It covers the fundamental components of seq2seq models including embeddings, encoders, decoders, attention mechanisms, and loss computation. The tutorial progresses from basic NMT implementations to attention-based models, includes hands-on training and inference sections, and covers practical optimizations like bidirectional RNNs, beam search, and multi-GPU training.