Large Language Model Recipes: A Hands-On Guide to Fine-Tuning, Optimization, Deployment, and Real-World Applications
暫譯: 大型語言模型食譜：實用指南於微調、優化、部署及實際應用

Bolla, Bharath Kumar, Subbaiah, Kalpa, Kaata, Sashi Kiran

Large Language Model Recipes: A Hands-On Guide to Fine-Tuning, Optimization, Deployment, and Real-World Applications
  • 出版商: Apress
  • 出版日期: 2026-06-19
  • 售價: $2,110
  • 貴賓價: 9.5 折 $2,004
  • 語言: 英文
  • 頁數: 402
  • 裝訂: Quality Paper - also called trade paper
  • ISBN: 9798868826061
  • ISBN-13: 9798868826061
  • 相關分類: Large language model

    • 海外代購書籍(需單獨結帳)

相關主題

商品描述

The Large Language Model Recipes book is a comprehensive, practical guide designed to help developers, data scientists, and AI engineers navigate the rapidly evolving landscape of Large Language Models (LLMs). Moving beyond theory, this book provides a hands-on, recipe-based approach to mastering the entire LLMs lifecycle, from selecting the right open-source model to fine-tuning it on custom data and deploying it for production at scale.

Starting with the fundamentals of setting up a robust development environment, the book guides you through the critical decisions of model selection (Llama, Mistral, Falcon) and data preparation. It offers deep dives into advanced training techniques, including full fine-tuning, instruction tuning, and parameter-efficient methods like LoRA and QLoRA that make training accessible on consumer hardware.

The book doesn't stop at training. It tackles the crucial "last mile" of AI development: deployment and optimization. You will learn how to shrink models with quantization, serve them with high-throughput engines like vLLM and TGI, and evaluate their performance using industry-standard benchmarks. Finally, it explores cutting-edge frontiers, including Retrieval-Augmented Generation (RAG) for grounding models in real-time data, building multimodal vision-language applications, and designing autonomous AI agents.

Whether you are building a specialized chatbot, a code assistant, or a complex reasoning agent, this book provides the tested recipes and code you need to develop efficient, scalable, and robust AI solutions today.

What you will learn:

    Design production-ready LLM systems using the Feature/Training/Inference (FTI) framework Apply advanced fine-tuning methods, including LoRA and QLoRA, for efficient model adaptation Build and optimize RAG pipelines with effective retrieval strategies and vector databases Deploy optimized LLMs using quantization techniques and scalable inference frameworks Develop multimodal and agentic AI applications with vision-language models and autonomous agents

Who this book is for:

This book is ideal for software developers, machine learning engineers, data scientists, and technical researchers who want to move beyond using API endpoints and start

商品描述(中文翻譯)

《大型語言模型食譜》是一本全面且實用的指南，旨在幫助開發者、數據科學家和人工智慧工程師在快速變化的大型語言模型（Large Language Models, LLMs）領域中導航。這本書超越了理論，提供了一種基於實作的食譜方法，幫助讀者掌握整個LLMs的生命周期，從選擇合適的開源模型到在自定義數據上進行微調，並將其部署到生產環境中以實現大規模應用。

本書從建立穩健的開發環境的基本原則開始，指導讀者進行模型選擇（如Llama、Mistral、Falcon）和數據準備的關鍵決策。它深入探討了先進的訓練技術，包括全面微調、指令微調，以及像LoRA和QLoRA這樣的參數高效方法，使得在消費者硬體上進行訓練變得可行。

本書不僅限於訓練，還處理人工智慧開發中的關鍵“最後一公里”：部署和優化。您將學習如何通過量化縮小模型，使用高吞吐量引擎（如vLLM和TGI）提供服務，並使用行業標準基準評估其性能。最後，它探索了前沿技術，包括用於將模型與實時數據結合的檢索增強生成（Retrieval-Augmented Generation, RAG）、構建多模態視覺-語言應用程序，以及設計自主人工智慧代理。

無論您是在構建專門的聊天機器人、代碼助手，還是複雜的推理代理，本書都提供了經過驗證的食譜和代碼，幫助您今天開發高效、可擴展且穩健的人工智慧解決方案。

您將學到的內容：
- 使用特徵/訓練/推理（Feature/Training/Inference, FTI）框架設計生產就緒的LLM系統
- 應用先進的微調方法，包括LoRA和QLoRA，以實現高效的模型適應
- 構建和優化RAG管道，使用有效的檢索策略和向量數據庫
- 使用量化技術和可擴展的推理框架部署優化的LLMs
- 開發多模態和自主的人工智慧應用，使用視覺-語言模型和自主代理

本書適合對象：
本書非常適合希望超越使用API端點的軟體開發者、機器學習工程師、數據科學家和技術研究人員。

作者簡介

Bharath Kumar Bolla is a highly accomplished Data Science leader with over 15 years of experience, specializing in AI, NLP, and Deep Learning for the past decade. He holds an M.S. in Data Science (The University of Arizona) and an Executive MBA (Product Management).

As an Associate Director at Novartis, he currently drives strategic MLOps initiatives, successfully designing and scaling automated pipelines and deploying cutting-edge Generative AI solutions across multiple European markets. His commercial impact is notable, including architecting a Salesforce recommendation system (5-10% conversion boost) and developing an ML pricing optimization product (+$1.2M revenue at Verizon).

Beyond corporate leadership, Bharath is a prolific academic with over 30 peer-reviewed publications and multiple best paper awards. He is recognized as a "40 Under 40 Data Scientist" (2022) and an "AI Changemaker," and actively supports the community by reviewing AI books and mentoring students.

Kalpa Subbaiah is a leading Data Scientist and AI expert with over 17 years of experience, including more than a decade in Data Science and Machine Learning. She holds a Master's degree in Machine Learning and Artificial Intelligence from Liverpool John Moores University and specializes in building end-to-end AI solutions across Azure, Databricks, and AWS.

Kalpa is a GenAI expert, building production-grade LLM and RAG applications, fine-tuning models via Hugging Face, and architecting scalable multi-agent AI systems with robust evaluation frameworks. Her strong experience in finance and manufacturing drives projects in financial AI platforms, smart city solutions, and enterprise analytics, leveraging capabilities such as document intelligence and object detection.

As Vice President and Lead Data Scientist at JPMorgan Chase & Co., she drives large-scale AI/GenAI transformation across the enterprise. Highly certified (Azure Data Scientist/AI Engineer, AWS ML Specialist), she actively delivers global corporate and academic training as a technical trainer and mentor.

Sashi Kiran Kaata is a seasoned cloud data engineer, researcher, and technology leader with over 10 years of experience architecting large-scale data platforms and real-time analytics solutions. Holding a Master's degree in Information Science and Technology, he has core expertise in AWS, Snowflake, Databricks, and modern streaming frameworks. At First Citizens Bank, he led enterprise data modernization, designing resilient ingestion and governance architectures, including Data Movement Controls (DMC), which significantly improved platform performance, reliability, and regulatory compliance.

Sashi blends practical engineering with research-driven innovation, actively contributing to the community as a technical conference speaker on distributed systems and scalable cloud architectures. His work extends into blockchain-based workflow modernization, sustainable AI pipelines, and adaptive ETL systems designed to support the next generation of intelligent data platforms. He is a prolific author of numerous peer-reviewed publications on critical areas, including cloud cost optimization, self-healing systems, Green AI, and MicroLLMs.

作者簡介(中文翻譯)

Bharath Kumar Bolla 是一位成就卓越的數據科學領導者，擁有超過 15 年的經驗，專注於人工智慧 (AI)、自然語言處理 (NLP) 和深度學習，並在過去十年中積累了豐富的專業知識。他擁有亞利桑那大學的數據科學碩士學位以及產品管理的高階工商管理碩士學位。

作為諾華公司的副總監，他目前推動戰略性的 MLOps 計劃，成功設計和擴展自動化管道，並在多個歐洲市場部署尖端的生成式 AI 解決方案。他的商業影響顯著，包括架構一個 Salesforce 推薦系統（提升 5-10% 的轉換率）以及開發一個機器學習定價優化產品（在 Verizon 創造了超過 120 萬美元的收入）。

除了企業領導角色外，Bharath 還是一位多產的學者，擁有超過 30 篇的同行評審出版物和多個最佳論文獎。他被認可為「40 位 40 歲以下數據科學家」（2022 年）和「AI 改革者」，並積極支持社群，通過審閱 AI 書籍和指導學生來貢獻力量。

Kalpa Subbaiah 是一位領先的數據科學家和 AI 專家，擁有超過 17 年的經驗，其中包括超過十年的數據科學和機器學習經歷。她擁有利物浦約翰摩爾斯大學的機器學習和人工智慧碩士學位，專注於在 Azure、Databricks 和 AWS 上構建端到端的 AI 解決方案。

Kalpa 是一位生成式 AI 專家，構建生產級的 LLM 和 RAG 應用，通過 Hugging Face 進行模型微調，並設計可擴展的多代理 AI 系統，配備強大的評估框架。她在金融和製造領域的豐富經驗推動了金融 AI 平台、智慧城市解決方案和企業分析項目，利用文檔智能和物體檢測等能力。

作為摩根大通的副總裁和首席數據科學家，她推動企業內的大規模 AI/生成式 AI 轉型。她擁有多項認證（Azure 數據科學家/AI 工程師、AWS 機器學習專家），並作為技術培訓師和導師，積極提供全球企業和學術培訓。

Sashi Kiran Kaata 是一位經驗豐富的雲數據工程師、研究人員和技術領導者，擁有超過 10 年的經驗，專注於架構大規模數據平台和實時分析解決方案。他擁有信息科學與技術的碩士學位，並在 AWS、Snowflake、Databricks 和現代流式框架方面擁有核心專業知識。在 First Citizens Bank，他負責企業數據現代化，設計了彈性的數據攝取和治理架構，包括數據移動控制 (DMC)，顯著改善了平台的性能、可靠性和合規性。

Sashi 將實用工程與研究驅動的創新相結合，積極參與社群，作為技術會議的演講者，分享有關分佈式系統和可擴展雲架構的知識。他的工作延伸至基於區塊鏈的工作流程現代化、可持續的 AI 管道和設計用於支持下一代智能數據平台的自適應 ETL 系統。他是多篇同行評審出版物的多產作者，涵蓋雲成本優化、自我修復系統、綠色 AI 和微型 LLM 等關鍵領域。

類似商品