Master the Foundational Technologies of LLMs and Agents! "Modern Transformer: From Internal Model Structure to Multimodal/RAG/SLM Implementation and Optimization" to be Released July 2nd

Impress Corporation (Chiyoda-ku, Tokyo, President: Takashi Takahashi), which operates IT and design-related media businesses within the Impress Group, will release the book "Modern Transformer: From Internal Model Structure to Multimodal/RAG/SLM Implementation and Optimization" on July 2, 2026. This book comprehensively covers the internal structure and practical techniques of AI models using the latest technologies.

Unraveling the Internal Structure of Transformer, Supporting AI Evolution

With the rapid proliferation of generative AI such as ChatGPT and Gemini, a deep understanding of its foundational technology, "Transformer," has become extremely important for engineers and data scientists. Currently, we are at a major turning point with the continuous emergence of next-generation models that support multimodality, including language, and the acceleration of model miniaturization (SLM) and advanced agentification. However, merely using provided models superficially has limitations for advanced customization and solving real-world challenges. This book is a timely volume that allows readers to systematically learn the essence of modern Transformer models, from fundamentals to applications, as currently required.

From Mathematical Foundations to On-Site Implementation, All in One Go

A major feature of this book is that while it explains the Transformer structure in detail, it also includes abundant practical code examples using the latest libraries such as Hugging Face and vLLM. It covers architectural variations like decoder/encoder-only designs and Mixture-of-Experts (MoE), as well as text generation strategies (e.g., Tree of Thoughts), RAG (Retrieval-Augmented Generation), fine-tuning (e.g., QLoRA), and inference optimization (e.g., FlashAttention). By learning while actually operating and verifying the code, readers can acquire solid practical skills.

This Book is Recommended for the Following Individuals:

Those who want to deeply understand the internal structure of Transformer, the foundational technology of generative AI.

Those who want to learn implementation and optimization techniques for RAG, multimodal, and SLM.

Those interested in utilizing quantization techniques like QLoRA and scaling with FlashAttention.

Data scientists and machine learning engineers who want to enhance their practical expertise.

Those who want to simultaneously acquire theoretical and practical knowledge by operating on-site level code.

Sample Pages

Explaining the basics of architecture

Explaining multimodal and multimodal RAG

Chapter 2 of This Book Available for Free for Two Weeks

To commemorate the release of the book, Chapter 2 will be available for free for a limited two-week period. The content can be viewed on web browsers without registration using our proprietary "Impress Web Book Viewer," allowing access from PCs, smartphones, and other devices anytime, anywhere. The viewer also includes features for sharing book information on social media and checking purchase details.

- "Modern Transformer: From Internal Model Structure to Multimodal/RAG/SLM Implementation and Optimization"

Page with link to free preview: https://book.impress.co.jp/books/1125101146

[Availability Period: July 2, 2026 (Thu) to July 15, 2026 (Wed)]

Book Structure

◆ Part 1: Fundamentals of Modern Transformer Models

Chapter 1: Why is Transformer Necessary?

Chapter 2: Detailed Explanation of Transformer

◆ Part 2: Generative Transformer

Chapter 3: Model Families and Architectural Variations

Chapter 4: Text Generation Strategies and Prompting Techniques

Chapter 5: Preference Alignment and RAG

◆ Part 3: Specialized Models

Chapter 6: Multimodal Models

Chapter 7: Efficient Specialized SLMs

Chapter 8: LLM Training and Evaluation

Chapter 9: LLM Optimization and Scaling

Chapter 10: Ethical and Responsible LLMs

Bibliographic Information

Title: Modern Transformer: From Internal Model Structure to Multimodal/RAG/SLM Implementation and Optimization

Series: impress top gear

Author: Nicole Koenigstein (Author), QUEEP Inc. (Translator)

Release Date: July 2, 2026 (Thu)

Pages: 312 pages

Size: B5 variant

Price: 3,630 yen (3,300 yen + 10% tax)

E-book Price: 3,630 yen (3,300 yen + 10% tax) *Impress direct sales price

ISBN: 978-4-295-02446-0

◇ Amazon Book Information Page: https://www.amazon.co.jp/dp/4295024465/

◇ Impress Book Information Page: https://book.impress.co.jp/books/1125101146

Author Profile

Nicole Koenigstein

A renowned data scientist and quantitative researcher. She serves as Chief Data Scientist and Head of AI and Quantitative Analysis at the asset management firm Wieden Capital. She is an external expert for the European Commission regarding funding for LLM research and an external expert for the International Organization of Securities Commissions (IOSCO) to advise on the introduction of generative AI in regulated industries. Her previous book is "Math for Machine Learning" (Manning Publications).

Translator Profile

QUEEP Inc.

Provides services in computer system development, localization, and consulting. Recent translations include "Clean Architecture in Python: Crafting Reusable, Maintainable, and Testable Software" (published by Impress). Other translated works include "LLM Production System Construction Know-how: From Basics to Implementation, Operation Methods, and Application Construction Examples" and "Causal Inference and Causal Discovery with Python Libraries [Concepts and Practice]: Unlocking Causal Machine Learning" (all published by Impress).

About the impress top gear Series

The impress top gear series provides technical books that cover new trends required in the 21st-century IT era, as well as presenting existing technologies and knowledge in a contemporary context. The series aims to offer technologies and knowledge that will help readers shift into "top gear" in the IT field.

End

[Impress Corporation] https://www.impress.co.jp/

Impress Corporation is a company that comprehensively develops and operates consumer-oriented media such as the "Dekiru" series of PC instruction books, which has sold over 80 million copies worldwide; the "MdN" brand for design and graphics-related books; regular magazines like "Digital Camera Magazine"; the "Impress Watch series," one of Japan's largest digital comprehensive news services with high access rates; and enterprise-focused IT-related media including "IT Leaders." The company broadly engages in IT and design-related publishing media businesses, as well as digital media & services businesses.

[Impress Group] https://www.impressholdings.com/

Impress Holdings, Inc. (Chiyoda-ku, Tokyo, President: Yuki Tsukamoto) is a media group with Impress Holdings as its holding company. It develops highly specialized media & services and solutions businesses with core themes in "IT/Design," "Music," "Mountains/Nature," "Aviation/Railways," "Mobile Services," and "Academia/Science & Engineering." Furthermore, it is involved in the development and operation of content business platforms.