No. 496 of 500 No. 48 in Deep Learning

Transformers in Action

By Nicole Koenigstein

Published
November 25, 2025
Print length
256 pages
Reading level
Intermediate
ISBN-13
978-1633437883
Publisher
Manning Publications
Written for
Python developers and data scientists who want a current, hands-on understanding of transformer-based LLMs.

Synopsis

Transformers in Action, published by Manning in late 2025, explains the architecture behind ChatGPT, Gemini and Claude and shows how to adapt pretrained models for new tasks. Nicole Koenigstein organizes ten chapters into three parts. The first covers the original encoder-decoder transformer, self-attention and feed-forward layers. The second turns to decoder-only language models, including causal attention, key-value caching and mixture-of-experts designs, then to sampling parameters, prompting methods such as chain-of-thought, preference alignment with RLHF, DPO and GRPO, and variants of retrieval-augmented generation. The third addresses multimodal models, small specialized language models, hyperparameter search with Ray Tune and Optuna, parameter-efficient fine-tuning such as LoRA and QLoRA, deployment optimizations including vLLM and PagedAttention, and responsible use.

The mathematics is taught through executable Jupyter notebooks rather than standalone proofs, and Luis Serrano contributes the foreword. Readers gain a current, end-to-end picture of how modern LLMs are built, steered and served, with working code for Hugging Face models. Reviewer Sujit Pal noted that parameter-efficient fine-tuning receives few worked examples and that evaluating non-deterministic outputs is largely missing, so teams building production retrieval systems will need additional sources.

Scorecard

Editorial score

How it scored

Editorial score: 63.1 / 1005 weighted criteria

  • Influence30%

    13.5/30
  • Authority25%

    16.3/25
  • Practical value20%

    14.0/20
  • Currency15%

    12.8/15
  • Readability10%

    6.5/10
  • Editorial score

    63.1/100

How the rank was computed

Rank No. 496 of 500 comes from a composite of 46.2: observable signals 17.9 of 55 points plus editorial review 28.3 of 45 points.

All 11 inputs
InputPoints
Observable signals17.9/55
Reach (Amazon ratings volume)4.2/16.5
Consensus across best-of lists5.5/13.8
Awards and bestseller evidence0.0/8.3
Reader rating, sample-adjusted5.1/8.3
Amazon sales rank0.3/5.5
Recency2.8/2.8
Editorial review28.3/45
Author authority7.3/11.3
Influence6.1/13.5
Practical value6.3/9.0
Readability2.9/4.5
Currency5.7/6.8

Each criterion is scored from 1 to 10 and weighted, and the points add up to the editorial score. The rank combines observable signals (55 percent) with this score (45 percent). Read the methodology.

Verdict

A code-forward guide to transformers and LLM adaptation that covers wide ground in 256 pages, occasionally at the expense of depth.

Strengths

  • Current coverage through DPO, GRPO and vLLM
  • Math explained through runnable Jupyter notebooks
  • Broad scope in a compact volume

Keep in mind

Coverage of parameter-efficient fine-tuning and LLM evaluation is thin, and readers focused on BERT-style encoder models will find limited material.

Key takeaways

  1. Self-attention lets each token weigh every other token in its context, which is why transformers replaced recurrent networks for most language tasks.
  2. Decoder-only models generate text one token at a time, and techniques such as key-value caching and mixture-of-experts layers make that generation cheaper to scale.
  3. Preference alignment methods, from RLHF to DPO and GRPO, follow supervised fine-tuning and shape how a model responds rather than what it knows.
  4. Small specialized language models can handle narrow jobs such as classification or safety guardrails efficiently, lowering cost and allowing private deployment.

Notable facts

  • Luis Serrano wrote the foreword, and Manning's page carries endorsements from reviewers at Google, Samsung, Palo Alto Networks and Tangible AI. Source (opens in a new tab)

About the author

Nicole Koenigstein

Nicole Koenigstein co-founded the fintech company Quantmate, where she serves as chief AI officer, and works on LLM systems and multi-agent architectures for quantitative finance. She has served as an external expert for the European Commission on LLM research funding and for IOSCO on generative AI in regulated industries, and has given talks at Oxford, Cornell and J.P. Morgan.

Questions

What is Transformers in Action about?
Transformers in Action explains how the transformer architecture works and how to adapt pretrained large language models for your own tasks. It covers self-attention, decoder-only models, prompting, preference alignment with RLHF and DPO, retrieval-augmented generation, multimodal and small language models, fine-tuning with LoRA, deployment optimization and responsible use, with Jupyter notebooks demonstrating the math and code.
Is Transformers in Action worth reading in 2026?
Yes, for practitioners who want a current technical overview. Published in late 2025, it covers methods such as GRPO, DoRA, PagedAttention and vLLM that older transformer books omit. Its compact length means some topics, notably parameter-efficient fine-tuning and evaluating non-deterministic outputs, receive brief treatment, so production teams should pair it with more specialized resources.
Who should read Transformers in Action?
The book is aimed at Python developers, data scientists and machine learning engineers who already understand basic neural networks and want to work directly with transformer models. It suits readers who learn through code and notebooks. Beginners without programming experience should start with a gentler introduction, and readers focused on encoder models such as BERT may prefer a title devoted to them.

Sources

As an Amazon Associate, Best AI Books earns from qualifying purchases. Best AI Books is sponsored by Iternal.ai, whose founder and CEO, John Byron Hanby IV, wrote The AI Strategy Blueprint, our #1 ranked book. How we rank books.