Popular NLP book ships notebooks, plus a frank maintenance log

What it does

This repository holds the executable Jupyter notebooks for the O’Reilly book Natural Language Processing with Transformers. Each chapter maps to a notebook covering tasks like text classification, named entity recognition, summarization, and training transformers from scratch. The maintainers provide one-click launch badges for Colab, Kaggle, Gradient, and SageMaker Studio Lab, though most chapters expect a GPU to finish in reasonable time.

The interesting bit

The README is remarkably upfront about decay: it explicitly flags that TensorFlow sections are untested and break under Keras 3, and that Chapter 7’s question-answering stack is abandoned because its dependency tree is incompatible with Python 3.10. That honesty saves you from the usual archaeological dig through stale ML examples.

Key highlights

• Eleven chapter notebooks, from introduction to training transformers from scratch
• One-click launchers for four cloud GPU platforms (Colab, Kaggle, Gradient, Studio Lab)
• PyTorch code paths are actively maintained; TensorFlow paths are legacy-only
• Chapter 7 concepts (retriever + reader pipelines, dense passage retrieval) remain valid even though the runnable code is deprecated
• Covers production concerns like model compression and few-to-no-label learning

Caveats

• TensorFlow/Keras 2 code paths are no longer tested and break against current Keras 3
• Chapter 7 (Question Answering) is unmaintained due to an upstream deprecation chain (farm-haystack, Elasticsearch 7.x, pydantic 1.x)

Verdict

Worth bookmarking if you are working through the O’Reilly book or need a curated, chapter-by-chapter NLP transformers syllabus. Skip it if you are looking for a standalone, continuously updated tutorial repository; the bit rot is real and documented.