Step-by-step PyTorch tutorial building a character-level Recurrent Neural Network from scratch without high-level RNN wrappers.
Written by Sean Robertson, this classic PyTorch tutorial implements a recurrent network from raw linear layers (`i2h` and `i2o`), encodes Unicode strings into character tensors, steps hidden states explicitly across sequence time steps, and trains the network end-to-end. Building a recurrent loop manually makes the sequential hidden-state bottleneck of RNNs concrete before you move to parallel Transformer attention in Phase 5.
Course status is tracked independently from roadmap topic checkboxes so you can mark external courses without altering your topic mastery percentage.
Phase 4 Hands-On Coding Tutorial (Modules 4.4 & 4.5)
100% free official PyTorch tutorial with downloadable Python script and Colab notebook.
Status: Verified (2026-11-09)
Review these foundations before starting so you get the most out of this material:
In Phase 4: NLP & Language Modeling
Complete this coding tutorial during Modules 4.4 and 4.5 to experience building explicit recurrent hidden-state loops in PyTorch.
Unicode code points, UTF-8 byte streams, normalization forms, sentence segmentation, corpus cleaning, and vocabulary construction.
Autoregressive next-token prediction, conditional probability factorization, cross-entropy training, perplexity evaluation, and sampling.
Recurrent Neural Networks (RNNs), LSTMs, teacher forcing, seq2seq encoder-decoders, and the information & parallelization bottlenecks that led to Transformers.
A code-first introduction to neural networks covering automatic differentiation, backpropagation, language models, and GPT implementations.
Stanford's flagship graduate NLP curriculum covering word vectors (Word2Vec), neural language models, recurrent networks (RNN/LSTM), attention mechanisms, and pretrained Transformers.
Official 'Learn the Basics' tutorial series covering PyTorch Tensors, Datasets/DataLoaders, `nn.Module`, `torch.autograd`, optimization loops, and model checkpointing.