Packt · 2025
DeepSeek in Practice 《DeepSeek 实战》
From basics to fine-tuning, distillation, agent design, and prompt engineering of open source LLM 开源大语言模型的基础、微调、蒸馏、智能体设计与提示工程
Three parts, fifteen chapters. The first builds up what the model is: architecture, reasoning mechanics, and the prompting patterns the V and R series actually respond to. The second puts it to work through case studies in document understanding and financial document analysis, then agent workflows and the Model Context Protocol. The third covers distillation and deployment, including a two-stage fine-tune of Gemma 3 on CUAD for legal clause reasoning, and a framework for choosing an inference stack. 全书三部分、十五章。第一部分讲清模型本身：架构、推理机制，以及 V 系列与 R 系列真正吃得住的提示词写法。第二部分动手做，从文档理解和财报分析两个案例，到智能体工作流与 Model Context Protocol。第三部分讲蒸馏与部署，包括在 CUAD 上对 Gemma 3 做两阶段微调以处理法律条款推理，以及一套选择推理栈的判断框架。
Written for engineers who already work with Python, APIs, and tools like Ollama or llama.cpp, and who have to make decisions about an open weights model rather than read about one. 写给已经在用 Python、API，以及 Ollama、llama.cpp 这类工具的工程师：需要对一个开放权重模型作出取舍，而不只是读个热闹。
- Authors 作者 Andy Peng, Alex Strick van Linschoten, Duarte O. Carmo
- Publisher 出版社 Packt
- Published 出版时间 Nov 21, 2025 2025 年 11 月 21 日
- Edition 版次 1st 第 1 版
- Length 页数 478 pages 478 页
- Language 语言 English 英文
- ISBN-13 9781806020850