Copyright and edition

Inside DeepInfer: How an LLM Inference Engine Works

Except where otherwise noted, the book’s original text and illustrations, including the web and PDF editions, are licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International licence (CC BY-NC-SA 4.0):
https://creativecommons.org/licenses/by-nc-sa/4.0/.

You may share, translate and adapt this material for noncommercial purposes under the licence’s terms, with attribution, a link to the licence and an indication of changes. Shared adaptations must use the same or a compatible licence. For a separate commercial licence, contact the author. Exceptions and limitations to copyright remain unaffected. Separately licensed code and third-party material retain their own terms.

Version 1.0, October 2026. The companion code for this version is tagged v1.0 at
https://github.com/jzstark/inside-deepinfer-code.

This book was written with extensive assistance from AI models. The author directed its scope and structure, reviewed the text and checked its evidence. Responsibility for the final text and any remaining errors rests with the author.

This is an independent book. It is not affiliated with, sponsored by or endorsed by DeepSeek. Product and company names belong to their respective owners; this book uses them only to identify the software it describes.

The book describes DeepInfer at commit 774ce244b87cc0eb8d4b94ab9f5de2ab22049f01, inspected in October 2026. Later versions may differ. Line references, defaults and behaviour stated here apply to that edition.

DeepInfer is distributed under the Apache License 2.0. Source excerpts are reproduced under that licence with attribution. Common leading indentation has been removed for typesetting; the code is otherwise unchanged. Appendix C records the source edition, file hashes, excerpt ranges and licence notices. The companion code is also distributed under Apache-2.0; its reuse is governed by that licence.

The experiments are teaching models written for this book. None of them runs the DeepInfer engine; lab 16 executes one unmodified DeepInfer policy file on CPU tensors. Their outputs are not performance measurements of DeepInfer. Performance figures and service descriptions attributed to DeepSeek’s documentation or model report are DeepSeek’s reported findings; this book has not reproduced them.

This book is provided as is, without warranty. Check a claim against the source edition before relying on it.