Natural Language Processing with Transformers, Revised Edition

The Transformer Tree of Life
The Encoder Branch
The Decoder Branch
The Encoder-Decoder Branch
Conclusion

4. Multilingual Named Entity Recognition
The Dataset
Multilingual Transformers
A Closer Look at Tokenization
The Tokenizer Pipeline
The SentencePiece Tokenizer
Transformers for Named Entity Recognition
The Anatomy of the Transformers Model Class
Bodies and Heads
Creating a Custom Model for Token Classification
Loading a Custom Model
Tokenizing Texts for NER
Performance Measures
Fine-Tuning XLM-RoBERTa
Error Analysis
Cross-Lingual Transfer
When Does Zero-Shot Transfer Make Sense?
Fine-Tuning on Multiple Languages at Once
Interacting with Model Widgets
Conclusion

5. Text Generation
The Challenge with Generating Coherent Text
Greedy Search Decoding
Beam Search Decoding
Sampling Methods
Top-k and Nucleus Sampling
Which Decoding Method Is Best?
Conclusion

6. Summarization
The CNN/DailyMail Dataset
Text Summarization Pipelines
Summarization Baseline
GPT-2
T5
BART
PEGASUS
Comparing Different Summaries
Measuring the Quality of Generated Text
BLEU
ROUGE
Evaluating PEGASUS on the CNN/DailyMail Dataset
Training a Summarization Model
Evaluating PEGASUS on SAMSum
Fine-Tuning PEGASUS
Generating Dialogue Summaries
Conclusion

7. Question Answering
Building a Review-Based QA System
The Dataset
Extracting Answers from Text
Using Haystack to Build a QA Pipeline
Improving Our QA Pipeline
Evaluating the Retriever
Evaluating the Reader
Domain Adaptation
Evaluating the Whole QA Pipeline
Going Beyond Extractive QA
Conclusion

8. Making Transformers Efficient in Production
Intent Detection as a Case Study
Creating a Performance Benchmark
Making Models Smaller via Knowledge Distillation
Knowledge Distillation for Fine-Tuning
Knowledge Distillation for Pretraining
Creating a Knowledge Distillation Trainer
Choosing a Good Student Initialization
Finding Good Hyperparameters with Optuna
Benchmarking Our Distilled Model
Making Models Faster with Quantization
Benchmarking Our Quantized Model
Optimizing Inference with ONNX and the ONNX Runtime
Making Models Sparser with Weight Pruning
Sparsity in Deep Neural Networks
Weight Pruning Methods
Conclusion

9. Dealing with Few to No Labels
Building a GitHub Issues Tagger
Getting the Data
Preparing the Data
Creating Training Sets
Creating Training Slices
Implementing a Naive Bayesline
Working with No Labeled Data
Working with a Few Labels
Data Augmentation
Using Embeddings as a Lookup Table
Fine-Tuning a Vanilla Transformer
In-Context and Few-Shot Learning with Prompts
Leveraging Unlabeled Data
Fine-Tuning a Language Model
Fine-Tuning a Classifier
Advanced Methods
Conclusion

10. Training Transformers from Scratch
Large Datasets and Where to Find Them
Challenges of Building a Large-Scale Corpus
Building a Custom Code Dataset
Working with Large Datasets
Adding Datasets to the Hugging Face Hub
Building a Tokenizer
The Tokenizer Model
Measuring Tokenizer Performance
A Tokenizer for Python
Training a Tokenizer
Saving a Custom Tokenizer on the Hub
Training a Model from Scratch
A Tale of Pretraining Objectives
Initializing the Model
Implementing the Dataloader
Defining the Training Loop
The Training Run
Results and Analysis
Conclusion

11. Future Directions
Scaling Transformers
Scaling Laws
Challenges with Scaling
Attention Please!
Sparse Attention
Linearized Attention
Going Beyond Text
Vision
Tables
Multimodal Transformers
Speech-to-Text
Vision and Text
Where to from Here?

Index
About the Authors