diff --git a/README.md b/README.md index 40ff9ee..ea5d576 100644 --- a/README.md +++ b/README.md @@ -1,8 +1,19 @@ # LLM Customization and Fine-Tuning: Adaptation, Distillation, and Alignment + + + +
+ *From a prompt to a fine-tuned, aligned, monitored assistant: every technique for adapting an open-weights LLM to your domain, on a single GPU, with reproducible numbers and the honesty to show when a technique does not win.* -![The whole-book arc as one diagram: a base model (Qwen3-4B) moves left to right through the adaptation spectrum with increasing depth (few-shot/RAG in chapter 4, LoRA/QLoRA in chapter 5, full SFT in chapter 6, distillation in chapter 7, DPO alignment in chapter 8) into a focused domain assistant that is cheaper, private, and monitored. The spectrum sits on a data foundation (chapter 3: quality, attribution, splits) and under an operations layer (chapter 9: monitor, drift, rollback, retrain), and every step runs on a single GPU with reproducible, measured results.](assets/book-hero.png) +**Available now in Manning's [Early Access Program (MEAP)](https://www.manning.com/books/llm-customization-and-fine-tuning):** read the chapters as they are released, run the code alongside them, and get the full ebook and print edition on publication. + +**Read it on Manning → [manning.com/books/llm-customization-and-fine-tuning](https://www.manning.com/books/llm-customization-and-fine-tuning)** + + +Book cover: LLM Customization and Fine-Tuning: Adaptation, Distillation, and Alignment (Manning, MEAP) +
Welcome to the code repository for **LLM Customization and Fine-Tuning: Adaptation, Distillation, and Alignment** (Manning Publications). @@ -10,7 +21,7 @@ This repository contains all the runnable code, data, and examples from the book It is a hands-on, cost-aware playbook for turning a general-purpose open-weights LLM into a focused, cheaper, privately hosted system that beats a generic API call on your domain, and for keeping it reliable in production. Every technique is backed by runnable code that reproduces the book's measured results on a single GPU. -**Book Publisher:** Manning Publications (forthcoming) +**Book Publisher:** Manning Publications, available now in [Early Access (MEAP)](https://www.manning.com/books/llm-customization-and-fine-tuning) **Repository:** > **Get started in minutes.** Jump to the **[Quick start](#quick-start)** (clone, install for your accelerator, smoke-test), then [`code/README.md`](code/README.md) for the full environment setup. Your first LoRA fine-tune (chapter 2) runs in **under 10 minutes on a single 12 GB GPU**, or on Apple Silicon. @@ -21,6 +32,8 @@ It is a hands-on, cost-aware playbook for turning a general-purpose open-weights This book is a practitioner's playbook for adapting large language models to specific use cases in a production setting. It covers the full customization spectrum, from prompting (chapter 4) through parameter-efficient fine-tuning (chapter 5), full supervised fine-tuning (chapter 6), distillation (chapter 7), preference optimisation (chapter 8), and the operational layer that keeps a fine-tuned model honest in production (chapter 9). By the last chapter you have taken one model, Qwen3-4B, from a base model through LoRA, full SFT, distillation, and DPO alignment, and into a drift-and-safety-monitored deployment, measuring cost and quality at each step. +![The whole-book arc as one diagram: a base model (Qwen3-4B) moves left to right through the adaptation spectrum with increasing depth (few-shot/RAG in chapter 4, LoRA/QLoRA in chapter 5, full SFT in chapter 6, distillation in chapter 7, DPO alignment in chapter 8) into a focused domain assistant that is cheaper, private, and monitored. The spectrum sits on a data foundation (chapter 3: quality, attribution, splits) and under an operations layer (chapter 9: monitor, drift, rollback, retrain), and every step runs on a single GPU with reproducible, measured results.](assets/book-hero.png) + ### The problem it solves Most teams reach for a general-purpose API and then hit a wall: it is expensive at scale, it sends data outside walls it should not cross, it is too slow for the latency budget, or it simply does not know the domain, the house terminology, or the internal tools. The fix is to adapt a model you control, but that opens a thicket of questions. Prompt, retrieve, or fine-tune? LoRA, full SFT, distillation, or alignment? What will it cost, what data do you need, and how do you keep it from regressing after launch? This book grew out of answering the same enterprise question over and over, "we have a use case and a GPU budget, which adaptation technique and how?", and turning the answer into one reproducible, honest reference: a cost-aware decision framework plus runnable reference implementations for every technique, instead of disconnected tutorials. diff --git a/assets/Bahree-Book-Cover-MEAP.jpg b/assets/Bahree-Book-Cover-MEAP.jpg new file mode 100644 index 0000000..cf011bd Binary files /dev/null and b/assets/Bahree-Book-Cover-MEAP.jpg differ